Base SAS 9.4 Procedures Guide High-Performance Procedures, Third Edition

Base SAS 9.
4
Procedures Guide
High-Performance Procedures
Third Edition
The correct bibliographic citation for this manual is as follows: SAS Institute Inc. 2014. Base SAS 9.4 Procedures Guide:
High-Performance Procedures, Third Edition. Cary, NC: SAS Institute Inc.
Base SAS 9.4 Procedures Guide: High-Performance Procedures, Third Edition
Copyright 2014, SAS Institute Inc., Cary, NC, USA
All rights reserved. Produced in the United States of America.
For a hard-copy book: No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by
any means, electronic, mechanical, photocopying, or otherwise, without the prior written permission of the publisher, SAS Institute
Inc.
For a Web download or e-book: Your use of this publication shall be governed by the terms established by the vendor at the time
you acquire this publication.
The scanning, uploading, and distribution of this book via the Internet or any other means without the permission of the publisher is
illegal and punishable by law. Please purchase only authorized electronic editions and do not participate in or encourage electronic
piracy of copyrighted materials. Your support of others rights is appreciated.
U.S. Government License Rights; Restricted Rights: The Software and its documentation is commercial computer software
developed at private expense and is provided with RESTRICTED RIGHTS to the United States Government. Use, duplication or
disclosure of the Software by the United States Government is subject to the license terms of this Agreement pursuant to, as
applicable, FAR 12.212, DFAR 227.7202-1(a), DFAR 227.7202-3(a) and DFAR 227.7202-4 and, to the extent required under U.S.
federal law, the minimum restricted rights as set out in FAR 52.227-19 (DEC 2007). If FAR 52.227-19 is applicable, this provision
serves as notice under clause (c) thereof and no other notice is required to be affixed to the Software or documentation. The
Governments rights in Software and documentation shall be only those set forth in this Agreement.
SAS Institute Inc., SAS Campus Drive, Cary, North Carolina 27513.
August 2014
SAS provides a complete selection of books and electronic products to help customers use SAS software to its fullest potential. For
more information about our offerings, visit support.sas.com/bookstore or call 1-800-727-3228.
SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the
USA and other countries. indicates USA registration.
Other brand and product names are trademarks of their respective companies.
Gain Greater Insight into Your

SAS Software with SAS Books.
Discover all that you need on your journey to knowledge and empowerment.
support.sas.com/bookstore
for additional books and resources.

SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. indicates USA registration. Other brand and product names are
trademarks of their respective companies. 2013 SAS Institute Inc. All rights reserved. S107969US.0613
iv
Contents
Chapter 1.
Chapter 2.
Chapter 3.
Chapter 4.
Chapter 5.
Chapter 6.
Chapter 7.
Chapter 8.
Chapter 9.
Chapter 10.
Whats New in Base SAS High-Performance Procedures 13.2

Introduction . . . . . . . . . . . . . . . . . .
Shared Concepts and Topics . . . . . . . . . . . .
The HPBIN Procedure . . . . . . . . . . . . . .
The HPCORR Procedure . . . . . . . . . . . . .
The HPDMDB Procedure . . . . . . . . . . . . .
The HPDS2 Procedure . . . . . . . . . . . . . .
The HPIMPUTE Procedure . . . . . . . . . . . .
The HPSAMPLE Procedure . . . . . . . . . . . .
The HPSUMMARY Procedure . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
5
9
41
67
83
95
109
119
131
Subject Index
161
Syntax Index
163
vi
Credits and Acknowledgments
Credits
Documentation
Editing
Anne Baxter, Ed Huddleston
Documentation Support
Tim Arnold, Melanie Gratton
Software
The SAS High-Performance Analytics Server was implemented by the following members of the development
staff. Program development includes design, programming, debugging, support, and documentation. In
the following list, the names of the developers who currently provide primary support are listed first; other
developers and previous developers are also listed.
HPBIN
HPCORR
HPDMDB
HPDS2
HPIMPUTE
HPSAMPLE
HPSUMMARY
High-performance computing foundation
High-performance analytics foundation
Numerical routines
Tao Wang, Taiping He

Charles Shorb
Scott Pope
Mark Freskos, Tom P. Weber, Oliver
Schabenberger
Tao Wang
Ye Liu
Scott Mebust, Gordon Keener
Steve E. Krueger
Robert Cohen, Georges H. Guirguis, Trevor
Kearney, Richard Knight, Gang Meng, Scott
Pope, Oliver Schabenberger, Charles Shorb,
Tom P. Weber
Georges H. Guirguis, Scott Pope
The following people contribute to the SAS High-Performance Analytics Server with their leadership and
support: Chris Bailey, David Pope, Oliver Schabenberger, Renee Sciortino, Jonathan Wexler.
2 F CONTENTS
Testing
Tim Carter, Enzo DAndreti, Alex Fang, Girija Gavankar, Dright Ho, Seungho Huh, Nilesh Jakhotiya, Cheryl
LeSaint, Catherine Lopes, Jim McKenzie, Jim Metcalf, Huiping Miao, Phil Mohr, Bengt Pederson, Kim
Sherrill, Weihua Shi, Stephanie Tysinger
Internationalization Testing
Feng Gao, Alex (Wenqi) He, David Li, Haiyong Rong, Bin Sun, Frank (Jidong) Wang, Lina Xu
Technical Support
Phil Gibbs
Acknowledgments
Many people make significant and continuing contributions to the development of SAS software products.
The final responsibility for the SAS System lies with SAS alone. We hope that you will always let us know
your opinions about the SAS System and its documentation. It is through your participation that SAS software
is continuously improved.
Chapter 1
Whats New in Base SAS High-Performance

Procedures 13.2
Contents
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Infrastructure Enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
GRIDDATASERVER= Option in the PERFORMANCE Statement No Longer Needed
GRIDMODE= Option in the PERFORMANCE Statement No Longer Needed . . . .
Multiple Data Access Modes Supported in a Single Procedure Step . . . . . . . . . .
Running Asymmetrically Alongside SAP HANA . . . . . . . . . . . . . . . . . . . .
Procedure Enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HPBIN Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HPCORR Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
HPSAMPLE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview
Infrastructure enhancements have been made and new functionality has been added in several procedures.
Infrastructure Enhancements
GRIDDATASERVER= Option in the PERFORMANCE Statement No Longer
Needed
When you run high-performance analytical procedures alongside databases, you no longer need to specify
the GRIDDATASERVER= option in the PERFORMANCE statement or set the GRIDDATASERVER
environment variable. Any values that you specify are ignored.
GRIDMODE= Option in the PERFORMANCE Statement No Longer Needed

With one exception, which is described in the GRIDMODE= section of the PERFORMANCE statement
documentation in Chapter 3, Shared Concepts and Topics, you no longer need to specify the GRIDMODE=
4 F Chapter 1: Whats New in Base SAS High-Performance Procedures 13.2
option in the PERFORMANCE statement or set the GRIDMODE environment variable. Any values that you
specify are ignored.
Multiple Data Access Modes Supported in a Single Procedure Step

The modes in which data sets are accessed are determined automatically for each data set that is used in a
single procedure step.
Running Asymmetrically Alongside SAP HANA

High-performance analytical procedures can now read data from and write data to SAP HANA asymmetrically.
Procedure Enhancements
HPBIN Procedure
The HPBIN procedure now supports quantile binning.
HPCORR Procedure
The HPCORR procedure now supports the WITH statement. The WITH statement lists variables with which
correlations of the VAR statement variables are to be computed.
HPSAMPLE Procedure
The HPSAMPLE procedure now supports a partitioning of the input data into three parts when you specify
SAMPPCT2= along with the SAMPPCT= and PARTITION options. PROC HPSAMPLE now also supports
up to four target variables for stratified sampling.
Chapter 2
Introduction
Contents
Overview of Base SAS High-Performance Procedures . . . . . . . . . . . . . . . . . . . .
About This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Typographical Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Options Used in Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Online Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
SAS Technical Support Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of Base SAS High-Performance Procedures

This book describes Base SAS high-performance utility procedures that are included when you install one or
more of the traditional SAS products that are shown in the following table:
Traditional Product
SAS High-Performance Analytics Product
SAS/STAT
SAS/ETS
SAS/OR
SAS High-Performance Forecasting
SAS Enterprise Miner
SAS Text Miner
SAS High-Performance Statistics

SAS High-Performance Econometrics
SAS High-Performance Optimization
SAS High-Performance Forecasting
SAS High-Performance Data Mining
SAS High-Performance Text Mining
You can run high-performance utility procedures in single-machine mode without licensing any of the SAS
High-Performance Analytics products that are listed in the preceding table. However, to run these procedures
in distributed mode, you must license at least one of the High-Performance Analytics products. For more
information about single-machine mode and distributed mode, see Chapter 3, Shared Concepts and Topics.
About This Book

This book assumes that you are familiar with Base SAS software and with the books SAS Language Reference:
Concepts and Base SAS Procedures Guide. It also assumes that you are familiar with basic SAS System
concepts, such as using the DATA step to create SAS data sets and using Base SAS procedures (such as the
PRINT and SORT procedures) to manipulate SAS data sets.
6 F Chapter 2: Introduction
Chapter Organization
This book is organized as follows:
Chapter 2, this chapter, provides an overview of high-performance utility procedures.
Chapter 3, Shared Concepts and Topics, describes the modes in which SAS high-performance utility
procedures can execute.
Subsequent chapters describe the high-performance utility procedures. These chapters appear in alphabetical
order by procedure name. Each chapter is organized as follows:
The Overview section provides a brief description of the functinality provided by the procedure.
The Getting Started section provides a quick introduction to the procedure through a simple example.
The Syntax section describes the SAS statements and options that control the procedure.
The Details section discusses methodology and other topics, such as ODS tables.
The Examples section contains examples that use the procedure.
The References section contains references for the methodology.
Typographical Conventions
This book uses several type styles for presenting information. The following list explains the meaning of the
typographical conventions used in this book:
roman
is the standard type style used for most text.
UPPERCASE ROMAN
is used for SAS statements, options, and other SAS language elements when
they appear in the text. However, you can enter these elements in your own SAS
programs in lowercase, uppercase, or a mixture of the two.
UPPERCASE BOLD
is used in the Syntax sections initial lists of SAS statements and options.
oblique
is used in the syntax definitions and in text to represent arguments for which you
supply a value.
VariableName
is used for the names of variables and data sets when they appear in the text.
bold
is used to for matrices and vectors.
italic
is used for terms that are defined in the text, for emphasis, and for references to
publications.
monospace
is used for example code. In most cases, this book uses lowercase type for SAS
code.
Options Used in Examples F 7
Options Used in Examples

The HTMLBLUE style is used to create the graphs and the HTML tables that appear in the online documentation. The PEARLJ style is used to create the PDF tables that appear in the documentation. A style template
controls stylistic elements such as colors, fonts, and presentation attributes. You can specify a style template
in an ODS destination statement as follows:
ods html style=HTMLBlue;
. . .
ods html close;
ods pdf style=PearlJ;
. . .
ods pdf close;
Most of the PDF tables are produced by using the following SAS System option:
options papersize=(6.5in 9in);
If you run the examples, you might get slightly different output. This is a function of the SAS System options
that are used and the precision that your computer uses for floating-point calculations.
Online Documentation
This documentation is available online with the SAS System. To access SAS High-Performance Statistics
software documentation from the SAS windowing environment, select Help from the main menu and then
select SAS Help and Documentation. On the Contents tab, expand the SAS Products, Base SAS, and
Base SAS Procedures Guide: High-Performance Procedures items. Then expand chapters and click on
sections. You can search the documentation by using the Search tab.
You can also access the documentation by going to http://support.sas.com/documentation.
SAS Technical Support Services

The SAS Technical Support staff is available to respond to problems and answer technical questions regarding
the use of the high- performance utility procedures. Go to http://support.sas.com/techsup for
more information.
Chapter 3
Shared Concepts and Topics

Contents
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
Processing Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
Single-Machine Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
Distributed Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
Controlling the Execution Mode with Environment Variables and Performance Statement Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
Determining Single-Machine Mode or Distributed Mode . . . . . . . . . . . . . . . .
12
Data Access Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
Single-Machine Data Access Mode . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
Distributed Data Access Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
Determining the Data Access Mode . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
Alongside-the-Database Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
Alongside-LASR Distributed Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
Running High-Performance Analytical Procedures Alongside a SAS LASR Analytic Server

in Distributed Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
Starting a SAS LASR Analytic Server Instance . . . . . . . . . . . . . . . . . . . . .
21
Associating a SAS Libref with the SAS LASR Analytic Server Instance . . . . . . . .
21
Running a High-Performance Analytical Procedure Alongside the SAS LASR Analytic Server Instance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
22
Terminating a SAS LASR Analytic Server Instance . . . . . . . . . . . . . . . . . . .
23
Alongside-LASR Distributed Execution on a Subset of the Appliance Nodes . . . . . . . . .
23
Running High-Performance Analytical Procedures in Asymmetric Mode . . . . . . . . . . .
23
Running in Symmetric Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
Running in Asymmetric Mode on One Appliance . . . . . . . . . . . . . . . . . . . .
25
Running in Asymmetric Mode on Distinct Appliances . . . . . . . . . . . . . . . . .
26
Alongside-HDFS Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
Alongside-HDFS Execution by Using the SASHDAT Engine . . . . . . . . . . . . .
29
Alongside-HDFS Execution by Using the Hadoop Engine . . . . . . . . . . . . . . .
31
Output Data Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
Working with Formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
PERFORMANCE Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
10 F Chapter 3: Shared Concepts and Topics
Overview
This chapter describes the modes of execution in which SAS high-performance analytical procedures can
execute. If you have Base SAS installed, you can run any procedure in this book on a single machine.
However, to run procedures in this book in distributed mode, you must also have SAS High-Performance
Server Distributed Mode software installed. For more information about these modes, see the next section.
This chapter provides details of how you can control the modes of execution and includes the syntax for the
PERFORMANCE statement, which is common to all high-performance analytical procedures.
Processing Modes
Single-Machine Mode
Single-machine mode is a computing model in which multiple processors or multiple cores are controlled
by a single operating system and can access shared resources, such as disks and memory. In this book,
single-machine mode refers to an application running multiple concurrent threads on a multicore machine
in order to take advantage of parallel execution on multiple processing units. More simply, single-machine
mode for high-performance analytical procedures means multithreading on the client machine.
All high-performance analytical procedures are capable of running in single-machine mode, and this is the
default mode when a procedure runs on the client machine. The procedure uses the number of CPUs (cores)
on the machine to determine the number of concurrent threads. High-performance analytical procedures use
different methods to map core count to the number of concurrent threads, depending on the analytic task.
Using one thread per core is not uncommon for the procedures that implement data-parallel algorithms.
Distributed Mode
Distributed mode is a computing model in which several nodes in a distributed computing environment
participate in the calculations. In this book, the distributed mode of a high-performance analytical procedure
refers to the procedure performing the analytics on an appliance that consists of a cluster of nodes. This
appliance can be one of the following:
a database management system (DBMS) appliance on which the SAS High-Performance Analytics
infrastructure is also installed
a cluster of nodes that have the SAS High-Performance Analytics infrastructure installed but no DBMS
software installed
Controlling the Execution Mode with Environment Variables and Performance Statement Options F 11
Controlling the Execution Mode with Environment Variables and

Performance Statement Options
You control the execution mode by using environment variables or by specifying options in the PERFORMANCE statement in high-performance analytical procedures, or by a combination of these methods.
The important environment variables follow:
grid host identifies the domain name system (DNS) or IP address of the appliance node to which the
SAS High-Performance Server Distributed Mode software connects to run in distributed mode.
installation location identifies the directory where the SAS High-Performance Server Distributed Mode
software is installed on the appliance.
You can set an environment variable directly from the SAS program by using the OPTION SET= command.
For example, the following statements define the grid host and the location where the SAS High-Performance
software is installed on the appliance:
option set=GRIDHOST
="hpa.sas.com";
option set=GRIDINSTALLLOC="/opt/TKGrid";
Alternatively, you can set the parameters in the PERFORMANCE statement in high-performance analytical
procedures. For example:
performance host
install
="hpa.sas.com"
="/opt/TKGrid";
A specification in the PERFORMANCE statement overrides a specification of an environment variable

without resetting its value. An environment variable that you set in the SAS session by using an OPTION
SET= command remains in effect until it is modified or until the SAS session terminates.
The key variable that determines whether a high-performance analytical procedure executes in single-machine
or distributed mode is the grid host. The installation location is needed to ensure that a connection to the grid
host can be made, given that a host is specified. This book assumes that the installation location has been set
by your system administrator.
The following sets of SAS statements are functionally equivalent:
proc hpreduce;
reduce unsupervised x:;
performance host="hpa.sas.com";
run;
option set=GRIDHOST="hpa.sas.com";
proc hpreduce;
reduce unsupervised x:;
run;
Determining Single-Machine Mode or Distributed Mode

High-performance analytical procedures use the following rules to determine whether they run in singlemachine mode or distributed mode:
If a grid host is not specified, the analysis is carried out in single-machine mode on the client machine
that runs the SAS session.
If a grid host is specified, the behavior depends on whether the execution is alongside the database
or alongside HDFS. If the data are local to the client (that is, not stored in the distributed database or
HDFS on the appliance), you need to use the NODES= option in the PERFORMANCE statement
to specify the number of nodes on the appliance or cluster that you want to engage in the analysis.
If the procedure executes alongside the database or alongside HDFS, you do not need to specify the
NODES= option.
The following example shows single-machine and client-data distributed configurations for a data set of
100,000 observations that are simulated from a logistic regression model. The following DATA step generates
the data:
data simData;
array _a{8} _temporary_ (0,0,0,1,0,1,1,1);
array _b{8} _temporary_ (0,0,1,0,1,0,1,1);
array _c{8} _temporary_ (0,1,0,0,1,1,0,1);
do obsno=1 to 100000;
x = rantbl(1,0.28,0.18,0.14,0.14,0.03,0.09,0.08,0.06);
a = _a{x};
b = _b{x};
c = _c{x};
x1 = int(ranuni(1)*400);
x2 = 52 + ranuni(1)*38;
x3 = ranuni(1)*12;
lp = 6. -0.015*(1-a) + 0.7*(1-b) + 0.6*(1-c) + 0.02*x1 -0.05*x2 - 0.1*x3;
y = ranbin(1,1,(1/(1+exp(lp))));
output;
end;
drop x lp;
run;
The following statements run PROC HPLOGISTIC to fit a logistic regression model:
proc hplogistic data=simData;

class a b c;
model y = a b c x1 x2 x3;
run;
Figure 3.1 shows the results from the analysis.
Determining Single-Machine Mode or Distributed Mode F 13
Figure 3.1 Results from Logistic Regression in Single-Machine Mode
The HPLOGISTIC Procedure

Performance Information
Execution Mode
Single-Machine
Number of Threads 4
Data Access Information
Data
Engine Role Path
WORK.SIMDATA V9
Input On Client
Model Information
Data Source
WORK.SIMDATA
Response Variable
Class Parameterization GLM

Distribution
Binary
Link Function
Logit
Optimization Technique Newton-Raphson with Ridging

Parameter Estimates
Parameter Estimate
Intercept
Standard
Error
DF t Value Pr > |t|
5.7011
0.2539 Infty
22.45 <.0001
a0
-0.01020
0.06627 Infty
-0.15 0.8777
a1
b0
0.7124
b1
c0
0.8036
c1
0.06558 Infty
.
0.06456 Infty
.
10.86 <.0001
.
12.45 <.0001
.
x1
0.01975 0.000614 Infty
32.15 <.0001
x2
-0.04728 0.003098 Infty
-15.26 <.0001
x3
-0.1017 0.009470 Infty
-10.74 <.0001
The entries in the Performance Information table show that the HPLOGISTIC procedure runs in singlemachine mode and uses four threads, which are chosen according to the number of CPUs on the client
machine. You can force a certain number of threads on any machine that is involved in the computations
by specifying the NTHREADS option in the PERFORMANCE statement. Another indication of execution
on the client is the following message, which is issued in the SAS log by all high-performance analytical
procedures:
NOTE: The HPLOGISTIC procedure is executing in single-machine mode.
The following statements use 10 nodes (in distributed mode) to analyze the data on the appliance; results
appear in Figure 3.2:

class a b c;
performance host="hpa.sas.com" nodes=10;
run;
Figure 3.2 Results from Logistic Regression in Distributed Mode

Host Node
hpa.sas.com
Execution Mode
Distributed
Number of Compute Nodes
10
Number of Threads per Node 24
The specification of a host causes the Performance Information table to display the name of the host node
of the appliance. The Performance Information table also indicates that the calculations were performed in
a distributed environment on the appliance. Twenty-four threads on each of 10 nodes were used to perform
the calculationsfor a total of 240 threads.
Another indication of distributed execution on the appliance is the following message, which is issued in the
SAS log by all high-performance analytical procedures:
NOTE: The HPLOGISTIC procedure is executing in the distributed

computing environment with 10 worker nodes.
You can override the presence of a grid host and force the computations into single-machine mode by
specifying the NODES=0 option in the PERFORMANCE statement:
class a b c;
run;
Figure 3.3 shows the Performance Information table. The numeric results are not reproduced here, but they
agree with the previous analyses, which are shown in Figure 3.1 and Figure 3.2.
Figure 3.3 Single-Machine Mode Despite Host Specification

Execution Mode
Single-Machine
Number of Threads 4
Data
Engine Role Path
WORK.SIMDATA V9
Input On Client
The Performance Information table indicates that the HPLOGISTIC procedure executes in single-machine
mode on the client. This information is also reported in the following message, which is issued in the SAS
log:
Data Access Modes F 15
NOTE: The HPLOGISTIC procedure is executing in single-machine mode.
In the analysis shown previously in Figure 3.2, the data set Work.simData is local to the client, and the
HPLOGISTIC procedure distributed the data to 10 nodes on the appliance. The High-Performance Analytics
infrastructure does not keep these data on the appliance. When the procedure terminates, the in-memory
representation of the input data on the appliance is freed.
When the input data set is large, the time that is spent sending client-side data to the appliance might dominate
the execution time. In practice, transfer speeds are usually lower than the theoretical limits of the network
connection or disk I/O rates. At a transfer rate of 40 megabytes per second, sending a 10-gigabyte data set
to the appliance requires more than four minutes. If analytic execution time is in the range of seconds, the
performance of the process is dominated by data movement.
The alongside-the-database execution model, unique to high-performance analytical procedures, enables you
to read and write data in distributed form from the database that is installed on the appliance.
Data Access Modes

Single-Machine Data Access Mode
When high-performance analytical procedures run in single-machine mode, they access data in the same
way as traditional SAS procedures. They use Base SAS to access input and output SAS data sets on the
client machine, and they use the relevant SAS/ACCESS interface to bring data from other sources, such as
third-party databases, Hadoop, and SAS LASR servers, to the client.
Distributed Data Access Mode

When high-performance analytical procedures run in distributed mode, input data must be brought to the
computation that is performed on the nodes of the grid, and output data must be sent from the computational
nodes. This can be accomplished in several ways:
Client-data (local-data) mode: The input and output data for the analytic task are stored on the client
machine where the high-performance procedure is invoked. When the procedure runs, the SAS HighPerformance Analytics infrastructure sends input data from the client to the distributed computing
environment and sends output data from the distributed computing environment to the client.
Parallel symmetric mode: Input and output data are stored on the same nodes that are used for the
distributed computation, and the data move in parallel from the data store to the computational nodes
without crossing node boundaries. Parallel symmetric mode is available with the following distributed
data sources:
Data in Greenplum or Teradata databases that are collocated with the computational nodes. This
access mode is also called alongside-the-database mode. For more information, see the section
Alongside-the-Database Execution on page 18.
Data in SASHDAT format in the Hadoop Distributed File System (HDFS) that is collocated
with the computational nodes. This access mode is also called alongside-HDFS mode. For more
information, see the section Alongside-HDFS Execution by Using the SASHDAT Engine on
page 29.
Data in a SAS LASR Analytic Server that is collocated with the computational nodes. This access
mode is also called alongside-LASR mode. For more information, see the section Running
High-Performance Analytical Procedures Alongside a SAS LASR Analytic Server in Distributed
Mode on page 20.
Parallel asymmetric mode: The primary reason for providing this mode is to enable you to manage and
house data on appliances (the data appliances) and to run high-performance analytical procedures on a
different appliance (the computing appliance). The high-performance analytical procedures run in a
SAS process on the computing appliance. For each data source that is accessed in parallel asymmetric
mode, a SAS Embedded Process must run on the associated data appliance. Data are requested by a
SAS data feeder that runs on the computing appliance and communicates with the SAS Embedded
Process on the data appliance. The SAS Embedded Process transfers the data in parallel to the SAS
data feeder that runs on each of the nodes of the computing appliance. This mode is called asymmetric
mode because the number of nodes on the data appliance does not need to match the number of nodes
on the computing appliance. Parallel asymmetric mode is supported for data in Teradata, Greenplum,
and Oracle databases and for data in HDFS and SAP HANA. In these cases, the parallel asymmetric
access is somewhat loosely described as being asymmetric alongside access, even though the data
storage and computation can occur on different appliances. For more information, see the section
Running High-Performance Analytical Procedures in Asymmetric Mode on page 23.
Through-the-client mode: When data can be accessed through a SAS/ACCESS interface but the data
reside in a file system or in a distributed data source on which a SAS Embedded Process is not running,
those data cannot be accessed in parallel in either symmetric or asymmetric mode. The SAS/ACCESS
interface is used to transfer input data from the data source to the client machine on which the highperformance procedure is invoked, and the data are then sent to the distributed computing environment
by the SAS High-Performance Analytics infrastructure. The data path is reversed for output data. This
mode of data access is referred to as through-the-client access.
Determining the Data Access Mode

High-performance analytical procedures determine the data access mode individually for each data set that
is used in the analysis. When high-performance analytical procedures run in distributed mode, parallel
symmetric or parallel asymmetric mode is used whenever possible. There are two reasons why parallel
access might not be possible. The first reason is that for a particular data set, the required SAS Embedded
Process is not installed on the appliance that houses the data. In such cases, access to those data reverts to
through-the-client access, and a note like the following is reported in the SAS log:
NOTE: The data MYLIB.MYDATA are being routed through the client because a
SAS Embedded Process is not running on the associated data server.
The second reason why parallel data access might not be possible for a particular data set is that the required
driver software might not be installed on the compute nodes. In this case, the required data feeder that
Determining the Data Access Mode F 17
moves the data from the compute nodes to the data source cannot be successfully loaded, and a note like the
following is reported in the SAS log:
NOTE: The data MYLIB.MYDATA are being routed through the client because
the ORACLE data feeder could not be loaded on the specified grid host.
For distributed data in SASHDAT format in HDFS or data in a SAS LASR Analytic Server, parallel symmetric
access is used when the data nodes and compute nodes are collocated on the same appliance. For data in a
LASR Analytic Server that cannot be accessed in parallel symmetric mode, through-the-client mode is used.
Through-the-client access is not supported for data in SASHDAT format in HDFS.
For data in Greenplum databases, parallel symmetric access is used if the compute nodes and the data nodes
are collocated on the same appliance and you do not specify the NODES=n option in a PERFORMANCE
statement. In this case, the number of nodes that are used is determined by the number of nodes across which
the data are distributed. If you specify NODES=n, then parallel asymmetric access is used.
For data in Teradata databases, parallel asymmetric mode is used by default. If your data and computation
are collocated on the same appliance and you want to use parallel symmetric mode, then you need to
specify GRIDMODE=SYM in the PERFORMANCE statement or you need to set the GRIDMODE=SYM
environment variable by using an OPTION SET statement.
High-performance analytical procedures produce a Data Access Information table that shows you how
each data set that is used in the analysis is accessed. The following statements provide an example in which
PROC HPDS2 is used to copy a distributed data set named Neuralgia (which is stored in SASHDAT format
in HDFS) to a SAS data set on the client machine:
libname hdatlib sashdat
host='hpa.sas.com';
hdfs_path="/user/hps";
proc hpds2 data=hdatlib.neuralgia out=neuralgia;
performance host='hpa.sas.com';
data DS2GTF.out;
method run();
set DS2GTF.in;
end;
enddata;
run;
Figure 3.4 shows the output that PROC HPDS2 produces. The Performance Information table shows that
PROC HPDS2 ran in distributed mode on a 13-node grid. The Data Access Information table shows that
the input data were accessed in parallel symmetric mode and the output data set was sent to the client, where
the V9 (base) engine stored it as a SAS data set in the Work directory.
Figure 3.4 Performance Information and Data Access Information Tables
The HPDS2 Procedure

Host Node
hpa.sas.com
Execution Mode
Distributed
13
Figure 3.4 continued

Data
Engine
Role
HDATLIB.NEURALGIA SASHDAT Input

WORK.NEURALGIA
V9
Path
Parallel, Symmetric
Output To Client
Alongside-the-Database Execution
High-performance analytical procedures interface with the distributed database management system (DBMS)
on the appliance in a unique way. If the input data are stored in the DBMS and the grid host is the appliance
that houses the data, high-performance analytical procedures create a distributed computing environment in
which an analytic process is collocated with the nodes of the DBMS. Data then pass from the DBMS to the
analytic process on each node. Instead of moving across the network and possibly back to the client machine,
the data pass locally between the processes on each node of the appliance.
Because the analytic processes on the appliance are separate from the database processes, the technique is
referred to as alongside-the-database execution in contrast to in-database execution, where the analytic code
executes in the database process.
In general, when you have a large amount of input data, you can achieve the best performance from
high-performance analytical procedures if execution is alongside the database.
Before you can run alongside the database, you must distribute the data to the appliance. The following
statements use the HPDS2 procedure to distribute the data set Work.simData into the mydb database on the
hpa.sas.com appliance. In this example, the appliance houses a Greenplum database.
option set=GRIDHOST="green.sas.com";
libname applianc greenplm
server ="green.sas.com"
user
=XXXXXX
password=YYYYY
database=mydb;
proc datasets lib=applianc nolist; delete simData;

proc hpds2 data=simData
out =applianc.simData(distributed_by='distributed randomly');
performance commit=10000 nodes=all;
data DS2GTF.out;
method run();
set DS2GTF.in;
end;
enddata;
run;
If the output table applianc.simData exists, the DATASETS procedure removes the table from the Greenplum
database because a DBMS does not usually support replacement operations on tables.
Alongside-the-Database Execution F 19
Note that the libref for the output table points to the appliance. The data set option informs the HPDS2
procedure to distribute the records randomly among the data segments of the appliance. The statements that
follow the PERFORMANCE statement are the DS2 program that copies the input data to the output data
without further transformations.
Because you loaded the data into a database on the appliance, you can use the following HPLOGISTIC
statements to perform the analysis on the appliance in the alongside-the-database mode. These statements
are almost identical to the first PROC HPLOGISTIC example in a previous section, which executed in
single-machine mode.
proc hplogistic data=applianc.simData;

class a b c;
run;
The subtle differences are as follows:

The grid host environment variable that you specified in an OPTION SET= command is still in effect.
The DATA= option in the high-performance analytical procedure uses a libref that identifies the data
source as being housed on the appliance. This libref was specified in a prior LIBNAME statement.
Figure 3.5 shows the results from this analysis. The Performance Information table shows that the
execution was in distributed mode, and the Data Access Information table shows that the data were
accessed asymmetrically in parallel from the Greenplum database. The numeric results agree with the
previous analyses, which are shown in Figure 3.1 and Figure 3.2.
Figure 3.5 Alongside-the-Database Execution on Greenplum

Host Node
green.sas.com
Execution Mode
Distributed
16

Data
Engine
Role Path
APPLIANC.SIMDATA GREENPLM Input Parallel, Asymmetric

Model Information
Data Source
APPLIANC.SIMDATA
Response Variable

Distribution
Binary
Link Function
Logit

Parameter Estimates
Parameter Estimate
Intercept
Standard
Error
DF t Value Pr > |t|
5.7011
0.2539 Infty
22.45 <.0001
a0
-0.01020
0.06627 Infty
-0.15 0.8777
a1
b0
0.7124
b1
c0
0.8036
c1
0.06558 Infty
.
0.06456 Infty
.
10.86 <.0001
.
12.45 <.0001
.
x1
0.01975 0.000614 Infty
32.15 <.0001
x2
-0.04728 0.003098 Infty
-15.26 <.0001
x3
-0.1017 0.009470 Infty
-10.74 <.0001
Alongside-LASR Distributed Execution

You can execute high-performance analytical procedures in distributed mode alongside a SAS LASR Analytic
Server. When high-performance analytical procedures run in this mode, the data are preloaded in distributed
form in memory that is managed by a LASR Analytic Server. The data on the nodes of the appliance
are accessed in parallel in the process that runs the LASR Analytic Server, and they are transferred to the
process where the high-performance analytical procedure runs. In general, each high-performance analytical
procedure copies the data to memory that persists only while that procedure executes. Hence, when a
high-performance analytical procedure runs alongside a LASR Analytic Server, both the high-performance
analytical procedure and the LASR Analytic Server have a copy of the subset of the data that is used by the
high-performance analytical procedure. The advantage of running high-performance analytical procedures
alongside a LASR Analytic Server (as opposed to running alongside a DBMS table or alongside HDFS) is
that the initial transfer of data from the LASR Analytic Server to the high-performance analytical procedure
is a memory-to-memory operation that is faster than the disk-to-memory operation when the procedure runs
alongside a DBMS or HDFS. When the cost of preloading a table into a LASR Analytic Server is amortized
by multiple uses of these data in separate runs of high-performance analytical procedures, using the LASR
Analytic Server can result in improved performance.
Running High-Performance Analytical Procedures Alongside

a SAS LASR Analytic Server in Distributed Mode
This section provides an example of steps that you can use to start and load data into a SAS LASR Analytic
Server instance and then run high-performance analytical procedures alongside this LASR Analytic Server
instance.
Associating a SAS Libref with the SAS LASR Analytic Server Instance F 21
Starting a SAS LASR Analytic Server Instance

The following statements create a SAS LASR Analytic Server instance and load it with the simData data
set that is used in the preceding examples. The data that are loaded into the LASR Analytic Server persist
in memory across procedure boundaries until these data are explicitly deleted or until the server instance is
terminated.
proc lasr port=54545

data=simData
path="/tmp/";
performance host="hpa.sas.com" nodes=ALL;
run;
The PORT= option specifies a network port number to use. The PATH= option specifies the directory in
which the server and table signature files are to be stored. The specified directory must exist on each machine
in the cluster. The DATA= option specifies the name of a data set that is loaded into this LASR Analytic
Server instance. (You do not need to specify the DATA= option at this time because you can add tables to
the LASR Analytic Server instance at any stage of its life.) For more information about starting and using a
LASR Analytic Server, see the SAS LASR Analytic Server: Reference Guide.
The NODES=ALL option in the PERFORMANCE statement specifies that the LASR Analytic Server run
on all the nodes on the appliance. You can start a LASR Analytic Server on a subset of the nodes on an
appliance, but this might affect whether high-performance analytical procedures can run alongside the LASR
Analytic Server. For more information, see the section Alongside-LASR Distributed Execution on a Subset
of the Appliance Nodes on page 23.
Figure 3.6 shows the Performance Information and Data Access Information tables, which show that
the LASR procedure ran in distributed mode on 13 nodes and that the data were sent from the client to the
appliance.
Figure 3.6 Performance and Data Access Information
The LASR Procedure

Host Node
hpa.sas.com
Execution Mode
Distributed
Number of Compute Nodes 13

Data
Engine Role Path
WORK.SIMDATA V9
Input From Client
Associating a SAS Libref with the SAS LASR Analytic Server Instance
The following statements use a LIBNAME statement that associates a SAS libref (named MyLasr) with
tables on the server instance as follows:
libname MyLasr sasiola port=54545 host="hpa.sas.com";
The SASIOLA option requests that the MyLasr libref use the SASIOLA engine, and the PORT= value
associates this libref with the appropriate server instance. For more information about creating a libref that
uses the SASIOLA engine, see the SAS LASR Analytic Server: Reference Guide.
Running a High-Performance Analytical Procedure Alongside the SAS

LASR Analytic Server Instance
You can use the MyLasr libref to specify the input data for high-performance analytical procedures. You can
also create output data sets in the SAS LASR Analytic Server instance by using this libref to request that the
output data set be held in memory by the server instance as follows:
proc hplogistic data=MyLasr.simData;

class a b c;
output out=MyLasr.simulateScores pred=PredictedProbabliity;
run;
Because you previously specified the GRIDHOST= environment variable and the input data are held in
distributed form in the associated server instance, this PROC HPLOGISTIC step runs in distributed mode
alongside the LASR Analytic Server, as indicated in the Performance Information table shown in Figure 3.7.
Figure 3.7 Performance and Data Access Information

Host Node
hpa.sas.com
Execution Mode
Distributed
13

Data
Engine
MYLASR.SIMDATA
SASIOLA Input
Role
Path
Parallel, Symmetric
MYLASR.SIMULATESCORES SASIOLA Output Parallel, Symmetric
The Data Access Information table shows that both the input and output data were read and written,
respectively, in parallel symmetric mode.
The preceding OUTPUT statement creates an output table that is added to the LASR Analytic Server instance.
Output data sets do not have to be created in the same server instance that holds the input data. You can use a
different LASR Analytic Server instance to hold the output data set. However, in order for the output data to
be created in parallel symmetric mode, all the nodes that are used by the server instance that holds the input
data must also be used by the server instance that holds the output data.
Terminating a SAS LASR Analytic Server Instance F 23
Terminating a SAS LASR Analytic Server Instance

You can continue to run high-performance analytical procedures and add and delete tables from the SAS
LASR Analytic Server instance until you terminate the server instance as follows:
proc lasr term port=54545;

run;
Alongside-LASR Distributed Execution on a Subset of the

Appliance Nodes
When you run PROC LASR to start a SAS LASR Analytic Server, you can specify the NODES= option in a
PERFORMANCE statement to control how many nodes the LASR Analytic Server executes on. Similarly,
a high-performance analytical procedure can execute on a subset of the nodes either because you specify
the NODES= option in a PERFORMANCE statement or because you run alongside a DBMS or HDFS
with an input data set that is distributed on a subset of the nodes on an appliance. In such situations, if a
high-performance analytical procedure uses nodes on which the LASR Analytic Server is not running, then
running alongside LASR is not supported. You can avoid this issue by specifying the NODES=ALL in the
PERFORMANCE statement when you use PROC LASR to start the LASR Analytic Server.
Running High-Performance Analytical Procedures in

Asymmetric Mode
This section provides examples of how you can run high-performance analytical procedures in asymmetric
mode. It also includes examples that run in symmetric mode to highlight differences between the modes. For
a description of asymmetric mode, see the section Distributed Data Access Mode on page 15.
Asymmetric mode is commonly used when the data appliance and the computing appliance are distinct
appliances. In order to be able to use an appliance as a data provider for high-performance analytical
procedures that run in asymmetric mode on another appliance, it is not necessary that SAS High-Performance
Server Distributed Mode be installed on the data appliance. However, it is essential that a SAS Embedded
Process be installed on the data appliance and that SAS High-Performance Server Distributed Mode be
installed on the computing appliance.
The following examples use a 24-node data appliance named data_appliance.sas.com, which houses a
Teradata DBMS and has a SAS Embedded Process installed. Because SAS High-Performance Server
Distributed Mode is also installed on this appliance, it can be used to run high-performance analytical
procedures in both symmetric and asymmetric modes.
The following statements load the simData data set of the preceding sections onto the data appliance:
libname dataLib teradata

server ="tera2650"
user
=XXXXXX
password=YYYYY
database=mydb;
data dataLib.simData;
set simData;
run;
N OTE : You can provision the appliance with data even if SAS High-Performance Server Distributed Mode
software is not installed on the appliance.
The following subsections show how you can run the HPLOGISTIC procedure symmetrically and asymmetrically on a single data appliance and asymmetrically on distinct data and computing appliances.
Running in Symmetric Mode

The following statements run the HPLOGISTIC procedure in symmetric mode on the data appliance:
proc hplogistic data=dataLib.simData;

class a b c;
performance host
= "data_appliance.sas.com"
nodes
= 10
gridmode
= sym;
run;
Figure 3.8 shows the results of this analysis.

Figure 3.8 Alongside-the-Database Execution in Symmetric Mode on Teradata

Host Node
data_appliance.sas.com
Execution Mode
Distributed
24

Data
Engine
Role Path
DATALIB.simData TERADATA Input Parallel, Symmetric
Running in Asymmetric Mode on One Appliance F 25

Model Information
Data Source
DATALIB.simData
Response Variable

Distribution
Binary
Link Function
Logit

Parameter Estimates
Parameter Estimate
Intercept
Standard
Error
DF t Value Pr > |t|
5.7011
0.2539 Infty
22.45 <.0001
a0
-0.01020
0.06627 Infty
-0.15 0.8777
a1
b0
0.7124
b1
c0
0.8036
c1
0.06558 Infty
.
10.86 <.0001
0.06456 Infty
.
.
.
12.45 <.0001
x1
0.01975 0.000614 Infty
32.15 <.0001
x2
-0.04728 0.003098 Infty
-15.26 <.0001
x3
-0.1017 0.009470 Infty
-10.74 <.0001
The Performance Information table shows that the execution occurs in symmetric mode on the 24 nodes of
the data appliance. In this case, the NODES=10 option in the PERFORMANCE statement is ignored because
the number of nodes that are used is determined by the number of nodes across which the data are distributed,
as indicated in the following warning message in the SAS log:
WARNING: The NODES=10 option in the PERFORMANCE statement is ignored because

you are running alongside the distributed data source
DATALIB.simData.DATA. The number of compute nodes is determined by the
configuration of the distributed DBMS.
Running in Asymmetric Mode on One Appliance

By default, the HPLOGISTIC procedure runs in asymmetric mode, as shown Figure 3.9, which is produced
by the following statements:

class a b c;
performance host
= "data_appliance.sas.com"
nodes
= 10;
run;
Figure 3.9 Alongside-Teradata Execution in Asymmetric Mode

Host Node
data_appliance.sas.com
Execution Mode
Distributed
10

Data
Engine
Role Path
DATALIB.simData TERADATA Input Parallel, Asymmetric
The Performance Information table confirms that the NODES=10 option that you specified in the PERFORMANCE statement was honored, and the Data Access Information table shows that the data were
accessed in parallel asymmetric mode. The data were moved in parallel from the 24 nodes on which the data
were stored to the 10 nodes on which the execution occurred. The numeric results are not reproduced here,
but they agree with the previous analyses.
Running in Asymmetric Mode on Distinct Appliances

Usually, there is no advantage to executing high-performance analytical procedures in asymmetric mode
on one appliance, because data might have to be unnecessarily moved between nodes. The following
example demonstrates the more typical use of asymmetric mode. In this example, the specified grid host
compute_appliance.sas.com is a 142-node computing appliance that is different from the 24-node data
appliance data_appliance.sas.com, which houses the Teradata DBMS where the data reside.
The advantage of using different computing and data appliances is that the data appliance is not affected by
the execution of high-performance analytical procedures except during the initial parallel data transfer. A
potential disadvantage of this asymmetric mode of execution is that the performance can be limited by the
bandwidth with which data can be moved between the appliances. However, because this data movement
takes place in parallel from the nodes of the data appliance to the nodes of the computing appliance, this
potential performance bottleneck can be overcome with appropriately provisioned hardware. The following
statements show how this is done:

class a b c;
performance host = "compute_appliance.sas.com" nodes=30;
run;
Figure 3.10 shows the Performance Information and Data Access Information tables.
Running in Asymmetric Mode on Distinct Appliances F 27
Figure 3.10 Asymmetric Mode with Distinct Data and Computing Appliances

Host Node
compute_appliance.sas.com
Execution Mode
Distributed
30

Data
Engine
Role Path
PROC HPLOGISTIC ran on 30 nodes of the computing appliance, even though the data were partitioned
across the 24 nodes of the data appliance. The numeric results are not reproduced here, but they agree with
the previous analyses shown in Figure 3.1 and Figure 3.2.
Every time you run a high-performance analytical procedure in asymmetric mode that uses different computing and data appliances, data are transferred between these appliances. If you plan to make repeated use of
the same data, then it might be advantageous to temporarily persist the data that you need on the computing
appliance. One way to persist the data is to store them as a table in a SAS LASR Analytic Server that runs on
the computing appliance. By running PROC LASR in asymmetric mode, you can load the data in parallel
from the data appliance nodes to the nodes on which the LASR Analytic Server runs on the computing
appliance. You can then use a LIBNAME statement that associates a SAS libref with tables on the LASR
Analytic Server. The following statements show how you do this:
proc lasr port=54345
data=dataLib.simData
path="/tmp/";
performance host ="compute_appliance.sas.com" nodes=30;
run;
libname MyLasr sasiola tag="dataLib" port=54345 host="compute_appliance.sas.com" ;
Figure 3.11 show the Performance Information and Data Access Information tables.
Figure 3.11 PROC LASR Running in Asymmetric Mode
The LASR Procedure

Host Node
Execution Mode
Distributed

Data
Engine
Role Path
By default, all the nodes on the computing appliance would be used. However, because NODES=30 was
specified in the PERFORMANCE statement, PROC LASR ran on only 30 nodes of the computing appliance.
The data were loaded asymmetrically in parallel from the 24 data appliance nodes to the 30 compute nodes
on which PROC LASR ran.
After the data are loaded into a LASR Analytic Server that runs on the computing appliance, you can run
high-performance analytical procedures alongside this LASR Analytic Server as shown by the following
statements:

class a b c;
output out=MyLasr.myOutputData pred=myPred;
performance host = "compute_appliance.sas.com";
run;
The following note, which appears in the SAS log, confirms that the output data set is created successfully:
NOTE: The table DATALIB.MYOUTPUTDATA has been added to the LASR Analytic Server
with port 54345. The Libname is MYLASR.
You can use the dataLib libref that you used to load the data onto the data appliance to create an output data
set on the data appliance.

class a b c;
output out=dataLib.myOutputData pred=myPred;
run;
The following note, which appears in the SAS log, confirms that the output data set is created successfully on
the data appliance:
NOTE: The data set DATALIB.myOutputData has 100000 observations and 1 variables.
When you run a high-performance analytical procedure on a computing appliance and either read data from
or write data to a different data appliance on which a SAS Embedded Process is running, the Read and Write
operations take place in parallel without any movement of data to and from the SAS client.
When you no longer need the data in the SAS LASR Analytic Server, you should terminate the server instance
as follows:
proc lasr term port=54345;

performance host="compute_appliance.sas.com";
run;
If you configured Hadoop on the computing appliance, then you can create output data tables that are stored
in the HDFS on the computing appliance. You can do this by using the SASHDAT engine as described in the
section Alongside-HDFS Execution on page 29.
Alongside-HDFS Execution F 29
Alongside-HDFS Execution
Running high-performance analytical procedures alongside HDFS shares many features with running alongside the database. You can execute high-performance analytical procedures alongside HDFS by using either
the SASHDAT engine or the Hadoop engine.
You use the SASHDAT engine to read and write data that are stored in HDFS in a proprietary SASHDAT
format. In SASHDAT format, metadata that describe the data in the Hadoop files are included with the
data. This enables you to access files in SASHDAT format without supplying any additional metadata.
Additionally, you can also use the SASHDAT engine to read data in CSV (comma-separated value) format,
but you need supply metadata that describe the contents of the CSV data. The SASHDAT engine provides
highly optimized access to data in HDFS that are stored in SASHDAT format.
The Hadoop engine reads data that are stored in various formats from HDFS and writes data to HDFS in
CSV format. This engine can use metadata that are stored in Hive, which is a data warehouse that supplies
metadata about data that are stored in Hadoop files. In addition, this engine can use metadata that you create
by using the HDMD procedure.
The following subsections provide details about using the SASHDAT and Hadoop engines to execute
high-performance analytical procedures alongside HDFS.
Alongside-HDFS Execution by Using the SASHDAT Engine

If the grid host is a cluster that houses data that have been distributed by using the SASHDAT engine, then
high-performance analytical procedures can analyze those data in the alongside-HDFS mode. The procedures
use the distributed computing environment in which an analytic process is collocated with the nodes of the
cluster. Data then pass from HDFS to the analytic process on each node of the cluster.
Before you can run a procedure alongside HDFS, you must distribute the data to the cluster. The following
statements use the SASHDAT engine to distribute to HDFS the simData data set that was used in the previous
two sections:
option set=GRIDHOST="hpa.sas.com";
libname hdatLib sashdat
path="/hps";
data hdatLib.simData (replace = yes) ;
set simData;
run;
In this example, the GRIDHOST is a cluster where the SAS Data in HDFS Engine is installed. If a data set that
is named simData already exists in the hps directory in HDFS, it is overwritten because the REPLACE=YES
data set option is specified. For more information about using this LIBNAME statement, see the section
LIBNAME Statement for the SAS Data in HDFS Engine in the SAS LASR Analytic Server: Reference
Guide.
The following HPLOGISTIC procedure statements perform the analysis in alongside-HDFS mode. These
statements are almost identical to the PROC HPLOGISTIC example in the previous two sections, which
executed in single-machine mode and alongside-the-database distributed mode, respectively.
Figure 3.12 shows the Performance Information and Data Access Information tables. You see that the
procedure ran in distributed mode and that the input data were read in parallel symmetric mode. The numeric
results shown in Figure 3.13 agree with the previous analyses shown in Figure 3.1, Figure 3.2, and Figure 3.5.
Figure 3.12 Alongside-HDFS Execution Performance Information

Host Node
hpa.sas.com
Execution Mode
Distributed
13

Data
Engine
Role Path
HDATLIB.SIMDATA SASHDAT Input Parallel, Symmetric
Figure 3.13 Alongside-HDFS Execution Model Information

Model Information
Data Source
HDATLIB.SIMDATA
Response Variable

Distribution
Binary
Link Function
Logit

Parameter Estimates
Parameter Estimate
Intercept
Standard
Error
DF t Value Pr > |t|
5.7011
0.2539 Infty
22.45 <.0001
a0
-0.01020
0.06627 Infty
-0.15 0.8777
a1
b0
0.7124
b1
c0
0.8036
c1
0.06558 Infty
.
0.06456 Infty
.
10.86 <.0001
.
12.45 <.0001
.
x1
0.01975 0.000614 Infty
32.15 <.0001
x2
-0.04728 0.003098 Infty
-15.26 <.0001
x3
-0.1017 0.009470 Infty
-10.74 <.0001
Alongside-HDFS Execution by Using the Hadoop Engine F 31
Alongside-HDFS Execution by Using the Hadoop Engine

The following LIBNAME statement sets up a libref that you can use to access data that are stored in HDFS
and have metadata in Hive:
libname hdoopLib hadoop

server
= "hpa.sas.com"
user
= XXXXX
password
= YYYYY
database
= myDB
config
= "demo.xml" ;
For more information about LIBNAME options available for the Hadoop engine, see the LIBNAME topic in
the Hadoop section of SAS/ACCESS for Relational Databases: Reference. The configuration file that you
specify in the CONFIG= option contains information that is needed to access the Hive server. It also contains
information that enables this configuration file to be used to access data in HDFS without using the Hive
server. This information can also be used to specify replication factors and block sizes that are used when the
engine writes data to HDFS.
The following DATA step uses the Hadoop engine to distribute to HDFS the simData data set that was used
in the previous sections. The engine creates metadata for the data set in Hive.
data hdoopLib.simData;
set simData;
run;
After you have loaded data or if you are accessing preexisting data in HDFS that have metadata in Hive,
you can access this data alongside HDFS by using high-performance analytical procedures. The following
HPLOGISTIC procedure statements perform the analysis in alongside-HDFS mode. These statements are
similar to the PROC HPLOGISTIC example in the previous sections.
proc hplogistic data=hdoopLib.simData;

class a b c;
run;
Figure 3.14 shows the Performance Information and Data Access Information tables. You see that
the procedure ran in distributed mode and that the input data were read in parallel asymmetric mode. The
numeric results shown in Figure 3.15 agree with the previous analyses.
Figure 3.14 Alongside-HDFS Execution by Using the Hadoop Engine

Host Node
Execution Mode
Distributed
142

Data
Engine
Role Path
GRIDLIB.SIMDATA HADOOP Input Parallel, Asymmetric
Figure 3.15 Alongside-HDFS Execution by Using the Hadoop Engine

Model Information
Data Source
GRIDLIB.SIMDATA
Response Variable

Distribution
Binary
Link Function
Logit

Parameter Estimates
Parameter Estimate
Intercept
Standard
Error
DF t Value Pr > |t|
5.7011
0.2539 Infty
22.45 <.0001
a0
-0.01020
0.06627 Infty
-0.15 0.8777
a1
b0
0.7124
b1
c0
0.8036
c1
0.06558 Infty
.
0.06456 Infty
.
10.86 <.0001
.
12.45 <.0001
.
x1
0.01975 0.000614 Infty
32.15 <.0001
x2
-0.04728 0.003098 Infty
-15.26 <.0001
x3
-0.1017 0.009470 Infty
-10.74 <.0001
The Hadoop engine also enables you to access tables in HDFS that are stored in various formats and that are
not registered in Hive. You can use the HDMD procedure to generate metadata for tables that are stored in
the following file formats:
delimited text
fixed-record length binary
sequence files
XML text
To read any other kind of file in Hadoop, you can write a custom file reader plug-in in Java for use with
PROC HDMD. For more information about LIBNAME options available for the Hadoop engine, see the
LIBNAME topic in the Hadoop section of SAS/ACCESS for Relational Databases: Reference.
The following example shows how you can use PROC HDMD to register metadata for CSV data independently
from Hive and then analyze these data by using high-performance analytical procedures. The CSV data in the
Alongside-HDFS Execution by Using the Hadoop Engine F 33
table csvExample.csv is stored in HDFS in the directory /user/demo/data. Each record in this table consists
of the following fields, in the order shown and separated by commas.
1. a string of at most six characters
2. a numeric field with values of 0 or 1
3. a numeric field with real numbers
Suppose you want to fit a logistic regression model to these data, where the second field represents a target
variable named Success, the third field represents a regressor named Dose, and the first field represents a
classification variable named Group.
The first step is to use PROC HDMD to create metadata that are needed to interpret the table, as in the
following statements:
libname hdoopLib hadoop

server
= "hpa.sas.com"
user
= XXXXX
password
= YYYYY
HDFS_PERMDIR = "/user/demo/data"
HDFS_METADIR = "/user/demo/meta"
config
= "demo.xml"
DBCREATE_TABLE_EXTERNAL=YES;
proc hdmd name=hdoopLib.csvExample data_file='csvExample.csv'
format=delimited encoding=utf8 sep = ',';
column Group
column Success
column Dose
char(6);
double;
double;
run;
The metadata that are created by PROC HDMD for this table are stored in the directory /user/demo/meta
that you specified in the HDFS_METADIR = option in the preceding LIBNAME statement. After you create
the metadata, you can execute high-performance analytical procedures with these data by using the hdoopLib
libref. For example, the following statements fit a logistic regression model to the CSV data that are stored in
csvExample.csv table.
proc hplogistic data=hdoopLib.csvExample;
class Group;
model Success = Dose;
performance host
= "compute_appliance.sas.com"
gridmode = asym;
run;
Figure 3.16 shows the results of this analysis. You see that the procedure ran in distributed mode and that
the input data were read in parallel asymmetric mode. The metadata that you created by using the HDMD
procedure have been used successfully in executing this analysis.
Figure 3.16 Alongside-HDFS Execution with CSV Data

Host Node
Execution Mode
Distributed
142

Data
Engine
Role Path
GRIDLIB.CSVEXAMPLE HADOOP Input Parallel, Asymmetric

Model Information
Data Source
GRIDLIB.CSVEXAMPLE
Response Variable
Success

Distribution
Binary
Link Function
Logit

Class Level Information
Class Levels Values
Group
3 group1 group2 group3
Number of Observations Read 1000

Number of Observations Used 1000
Parameter Estimates
Parameter Estimate
Intercept
Dose
Standard
Error
DF t Value Pr > |t|
0.1243
0.1295 Infty
0.96 0.3371
-0.2674
0.2216 Infty
-1.21 0.2277
Output Data Sets

In the alongside-the-database mode, the data are read in distributed form, minimizing data movement for
best performance. Similarly, when you write output data sets and a high-performance analytical procedure
executes in distributed mode, the data can be written in parallel into the database.
For example, in the following statements, the HPLOGISTIC procedure executes in distributed mode by using
eight nodes on the appliance to perform the logistic regression on work.simData:

class a b c;
id a;
Working with Formats F 35
output out=applianc.simData_out pred=p;

run;
The output data set applianc.simData_out is written in parallel into the database. Although the data are fed
on eight nodes, the database might distribute the data on more nodes.
When a high-performance analytical procedure executes in single-machine mode, all output objects are
created on the client. If the libref of the output data sets points to the appliance, the data are transferred to the
database on the appliance. This can lead to considerable performance degradation compared to execution in
distributed mode.
Many procedures in SAS software add the variables from the input data set when an observationwise output
data set is created. The assumption of high-performance analytical procedures is that the input data sets can
be large and contain many variables. For performance reasons, the output data set contains the following:
variables that are explicitly created by the statement
variables that are listed in the ID statement, as described in Chapter 4, Shared Statistical Concepts
(SAS/STAT Users Guide: High-Performance Procedures)
distribution keys or hash keys that are transferred from the input data set
Including this information enables you to add to the output data set information necessary for subsequent
SQL joins without copying the entire input data set to the output data set.
Working with Formats

You can use SAS formats and user-defined formats with high-performance analytical procedures as you can
with other procedures in the SAS System. However, because the analytic work is carried out in a distributed
environment and might depend on the formatted values of variables, some special handling can improve the
efficiency of work with formats.
High-performance analytical procedures examine the variables that are used in an analysis for association with
user-defined formats. Any user-defined formats that are found by a procedure are transmitted automatically
to the appliance. If you are running multiple high-performance analytical procedures in a SAS session and
the analysis variables depend on user-defined formats, you can preprocess the formats. This step involves
generating an XML stream (a file) of the formats and passing the stream to the high-performance analytical
procedures.
Suppose that the following formats are defined in your SAS program:
proc format;
value YesNo
value checkThis
value $cityChar
run;
1='Yes'
1='ThisisOne'
1='Portage'
0='No';
2='ThisisTwo';
2='Kinston';
The next group of SAS statements create the XML stream for the formats in the file Myfmt.xml, associate that
file with the file reference myxml, and pass the file reference with the FMTLIBXML= option in the PROC
HPLOGISTIC statement:
filename myxml 'Myfmt.xml';

libname myxml XML92 xmltype=sasfmt tagset=tagsets.XMLsuv;
proc format cntlout=myxml.allfmts;
run;
proc hplogistic data=six fmtlibxml=myxml;

class wheeze cit age;
format wheeze best4. cit $cityChar.;
model wheeze = cit age;
run;
Generation and destruction of the stream can be wrapped in convenience macros:
%macro Make_XMLStream(name=tempxml);
filename &name 'fmt.xml';
libname &name XML92 xmltype=sasfmt tagset=tagsets.XMLsuv;
proc format cntlout=&name..allfmts;
run;
%mend;
%macro Delete_XMLStream(fref);
%let rc=%sysfunc(fdelete(&fref));
%mend;
If you do not pass an XML stream to a high-performance analytical procedure that supports the
FMTLIBXML= option, the procedure generates an XML stream as needed when it is invoked.
PERFORMANCE Statement
PERFORMANCE < performance-options > ;
The PERFORMANCE statement defines performance parameters for multithreaded and distributed computing, passes variables that describe the distributed computing environment, and requests detailed results about
the performance characteristics of a high-performance analytical procedure.
You can also use the PERFORMANCE statement to control whether a high-performance analytical procedure
executes in single-machine or distributed mode.
You can specify the following performance-options in the PERFORMANCE statement:
COMMIT=n
requests that the high-performance analytical procedure write periodic updates to the SAS log when
observations are sent from the client to the appliance for distributed processing.
High-performance analytical procedures do not have to use input data that are stored on the appliance.
You can perform distributed computations regardless of the origin or format of the input data, provided
PERFORMANCE Statement F 37
that the data are in a format that can be read by the SAS System (for example, because a SAS/ACCESS
engine is available).
In the following example, the HPREG procedure performs LASSO variable selection where the input
data set is stored on the client:
proc hpreg data=work.one;

model y = x1-x500;
selection method=lasso;
performance nodes=10 host='mydca' commit=10000;
run;
In order to perform the work as requested using 10 nodes on the appliance, the data set Work.One
needs to be distributed to the appliance.
High-performance analytical procedures send the data in blocks to the appliance. Whenever the number
of observations sent exceeds an integer multiple of the COMMIT= size, a SAS log message is produced.
The message indicates the actual number of observations distributed, and not an integer multiple of the
COMMIT= size.
DETAILS
requests a table that shows a timing breakdown of the procedure steps.

GRIDHOST=name
HOST=name
specifies the name of the appliance host in single or double quotation marks. If this option is specified,
it overrides the value of the GRIDHOST environment variable.
GRIDMODE=SYM | ASYM
MODE=SYM | ASYM
is a deprecated option that specifies whether to run the high-performance analytical procedure in
symmetric (SYM) mode or asymmetric (ASYM) mode. This option overrides the GRIDMODE
environment variable.
The only time you need to specify this option is when you want to run in symmetric mode alongside a
Teradata database. In all other cases, this option is not needed and the value that you specify is ignored.
GRIDTIMEOUT=s
TIMEOUT=s
specifies the time-out in seconds for a high-performance analytical procedure to wait for a connection
to the appliance and establish a connection back to the client. The default is 120 seconds. If jobs
are submitted to the appliance through workload management tools that might suspend access to the
appliance for a longer period, you might want to increase the time-out value.
INSTALL=name
INSTALLLOC=name
specifies the directory in which the shared libraries for the high-performance analytical procedure
are installed on the appliance. Specifying the INSTALL= option overrides the GRIDINSTALLLOC
environment variable.
LASRSERVER=path
LASR=path
specifies the fully qualified path to the description file of a SAS LASR Analytic Server instance. If
the input data set is held in memory by this LASR Analytic Server instance, then the procedure runs
alongside LASR. This option is not needed to run alongside LASR if the DATA= specification of the
input data uses a libref that is associated with a LASR Analytic Server instance. For more information,
see the section Alongside-LASR Distributed Execution on page 20 and the SAS LASR Analytic
Server: Reference Guide.
NODES=ALL | n
NNODES=ALL | n
specifies the number of nodes in the distributed computing environment, provided that the data are not
processed alongside the database.
Specifying NODES=0 indicates that you want to process the data in single-machine mode on the client
machine. If the input data are not alongside the database, this is the default. The high-performance
analytical procedures then perform the analysis on the client. For example, the following sets of
statements are equivalent:
proc hplogistic data=one;

model y = x;
run;

model y = x;
performance nodes=0;
run;
If the data are not read alongside the database, the NODES= option specifies the number of nodes
on the appliance that are involved in the analysis. For example, the following statements perform the
analysis in distributed mode by using 10 units of work on the appliance that is identified in the HOST=
option:

model y = x;
performance nodes=10 host="hpa.sas.com";
run;
If the number of nodes can be modified by the application, you can specify a NODES=n option,
where n exceeds the number of physical nodes on the appliance. The SAS High-Performance Server
Distributed Mode software then oversubscribes the nodes and associates nodes with multiple units of
work. For example, on a system that has 16 appliance nodes, the following statements oversubscribe
the system by a factor of 3:

model y = x;
run;
Usually, it is not advisable to oversubscribe the system because the analytic code is optimized for
a certain level of multithreading on the nodes that depends on the CPU count. You can specify
NODES=ALL if you want to use all available nodes on the appliance without oversubscribing the
system.
If the data are read alongside the distributed database on the appliance, specifying a nonzero value
for the NODES= option has no effect. The number of units of work in the distributed computing
environment is then determined by the distribution of the data and cannot be altered. For example, if
you are running alongside an appliance with 24 nodes, the NODES= option in the following statements
is ignored:
libname GPLib greenplm server=gpdca user=XXX password=YYY

database=ZZZ;
proc hplogistic data=gplib.one;
model y = x;
run;
NTHREADS=n
THREADS=n
specifies the number of threads for analytic computations and overrides the SAS system option
THREADS | NOTHREADS. If you do not specify the NTHREADS= option, the number of threads is
determined based on the number of CPUs on the host on which the analytic computations execute. The
algorithm by which a CPU count is converted to a thread count is specific to the high-performance
analytical procedure. Most procedures create one thread per CPU for the analytic computations.
By default, high-performance analytical procedures run in multiple concurrent threads unless multithreading has been turned off by the NOTHREADS system option or you force single-threaded
execution by specifying NTHREADS=1. The largest number that can be specified for n is 256. Individual high-performance analytical procedures can impose more stringent limits if called for by
algorithmic considerations.
N OTE : The SAS system options THREADS | NOTHREADS apply to the client machine on which the
SAS high-performance analytical procedures execute. They do not apply to the compute nodes in a
distributed environment.
40
Chapter 4
The HPBIN Procedure

Contents
Overview: HPBIN Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bucket Binning
42
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
Winsorized Binning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
Quantile Binning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
PseudoQuantile Binning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
PROC HPBIN Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
Getting Started: HPBIN Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
Syntax: HPBIN Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
46
PROC HPBIN Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
CODE Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
FREQ Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
ID Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
INPUT Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
PERFORMANCE Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
TARGET Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
50
Details: HPBIN Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
Computing the Quantiles (Percentiles) . . . . . . . . . . . . . . . . . . . . . . . . .
51
Binning Computation and Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
Computing the Weight of Evidence and Information Value . . . . . . . . . . . . . . .
54
Data Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
54
ODS Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
57
Example 4.1: Bucket Binning in Single-Machine Mode . . . . . . . . . . . . . . . .
57
Example 4.2: PseudoQuantile Binning in Distributed Mode . . . . . . . . . . . . .
59
Example 4.3: Quantile Binning in Distributed Mode . . . . . . . . . . . . . . . . . .
62
Example 4.4: Winsorized Binning . . . . . . . . . . . . . . . . . . . . . . . . . . .
64
Example 4.5: Bucket Binning and Weight-of-Evidence Computation . . . . . . . . .
65
42 F Chapter 4: The HPBIN Procedure
Overview: HPBIN Procedure

Binning is a common step in the data preparation stage of the model-building process. You can use binning
to classify missing variables, reduce the impact of outliers, and generate multiple effects. The generated
effects are useful and contain certain nonlinear information about the original interval variables.
The HPBIN procedure conducts high-performance binning by using binning methods that are described in
the following subsections. The HPBIN procedure can also calculate the weight of evidence (WOE) and
information value (IV) based on binning results.
The HPBIN procedure runs in either single-machine mode or distributed mode.
N OTE : Distributed mode requires SAS High-Performance Server Distributed Mode.
Bucket Binning
Bucket binning creates equal-length bins and assigns the data to one of these bins. You can choose the
number of bins during the binning; the default number of bins (the binning level) is 16.
Winsorized Binning
Winsorized binning is similar to bucket binning except that both tails are cut off to obtain a smooth binning
result. This technique is often used to remove outliers during the data preparation stage.
Quantile Binning
Quantile binning aims to assign the same number of observations to each bin, if the number of observations
is evenly divisible by the number of bins. As a result, each bin should have the same number of observations,
provided that there are no tied values at the boundaries of the bins. Because PROC HPBIN always assigns
observations that have the same value to the same bin, quantile binning might create unbalanced bins if any
variable has tied values. For more information, see the section Binning Computation and Formulas on
page 51.
PseudoQuantile Binning F 43
PseudoQuantile Binning
The HPBIN procedure offers pseudoquantile binning, which is an approximation of quantile binning. The
pseudoquantile binning method is very efficient, and the results mimic those of the quantile binning method.
PROC HPBIN Features

The HPBIN procedure has the following features:
provides a bucket (equal-length) binning method
provides a Winsorized binning method and Winsorized statistics
provides a quantile binning method and a pseudoquantile binning method
provides a mapping table for the selected binning method
provides a basic statistical table that contains the minimum, maximum, mean, median, and so on
computes the quantiles of the binning variables
calculates the weight of evidence (WOE) and information value (IV) based on binning results
reads input data in parallel and writes output data in parallel when the data source is in a database on
the appliance
Because the HPBIN procedure is a high-performance analytical procedure, it also does the following:
enables you to run in distributed mode on a cluster of machines that distribute the data and the
computations
enables you to run in single-machine mode on the server where SAS is installed
exploits all the available cores and concurrent threads, regardless of execution mode
For more information, see the section Processing Modes on page 10 in Chapter 3, Shared Concepts and
Topics.
Getting Started: HPBIN Procedure

This example shows how you can use the HPBIN procedure to perform pseudoquantile binning. Consider
the following statements:
data bindata;
do i=1 to 1000;
x=rannorm(1);
output;
end;
run;
proc rank data=bindata out=rankout group=8;
var x;
ranks rank_x;
run;
ods graphics on;
proc univariate data=rankout;
var rank_x;
histogram;
run;
These statements create a data set that contains 1,000 observations, each generated from a random normal
distribution. The histogram for this data set is shown in Figure 4.1.
Getting Started: HPBIN Procedure F 45
Figure 4.1 Histogram for Rank X
The pseudoquantile binning method in the HPBIN procedure can achieve a similar result by using far less
computation time. In this case, the time complexity is C O.n/, where C is a constant and n is the number
of observations. When the algorithm runs on the grid, the total amount of computation time is much less. For
example, if a cluster has 32 nodes and each node has 24 shared-memory CPUs, then the time complexity is
.C O.n//=.32 24/.
The following statements bin the data by using the PSEUDO_QUANTILE option in the PROC HPBIN
statement and generate the histogram for the binning output data. (See Figure 4.2.) This histogram is similar
to the one in Figure 4.1.
proc hpbin data=bindata output=binout numbin=8 pseudo_quantile;
input x;
run;
ods graphics on;
proc univariate data=binout;
var bin_x;
histogram;
run;
Figure 4.2 Histogram for Binning X
Syntax: HPBIN Procedure

The following statements are available in the HPBIN procedure:
PROC HPBIN < options > ;
CODE FILE=filename ;
FREQ variable ;
ID variables ;
INPUT variables < / option > ;
TARGET variable / LEVEL=level ORDER=order ;
The PROC HPBIN statement is required. You can specify multiple INPUT statements. The following sections
describe the PROC HPBIN statement and then describe the other statements in alphabetical order.
PROC HPBIN Statement F 47
PROC HPBIN Statement

PROC HPBIN < options > ;
The PROC HPBIN statement invokes the procedure. Table 4.1 summarizes important options in the PROC
HPBIN statement by function.
Table 4.1 PROC HPBIN Statement Options
Option
Description
Basic Options
DATA=
OUTPUT=
NOPRINT
Specifies the input data set

Specifies the output data set
Overwrites the ODS output
Binning Level Options

NUMBIN=
Specifies the global number of bins for all binning variables
Binning Method Options

BUCKET
PSEUDO_QUANTILE
QUANTILE
WINSOR
WINSORRATE=
Specifies the bucket binning method

Specifies the pseudoquantile binning method
Specifies the quantile binning method
Specifies the Winsorized binning method and the rate that it uses
Statistics Options
COMPUTESTATS
Computes the basic statistics of the binning variables
COMPUTEQUANTILE
Compute the quantiles of the binning variables
Weight-of-Evidence Options
BINS_META=
Specifies the BINS_META input data set, which contains the binning results
WOE
Computes the weight of evidence and information values
WOEADJUST=
Specifies the adjustment factor for the weight-of-evidence
calculation
You can specify the following optional arguments:

BINS_META=SAS-data-set
specifies the BINS_META input data set, which contains the binning results. The BINS_META data
set contains six variables: variable name, binned variable name, lower bound, upper bound, bin, and
range. The mapping table that is generated by PROC HPBIN can be used as the BINS_META data set.
BUCKET | QUANTILE | PSEUDO_QUANTILE | WINSOR WINSORRATE=number
specifies which binning method to use. If you specify BUCKET, then PROC HPBIN uses equallength binning. If you specify QUANTILE, then PROC HPBIN uses quantile binning. If you
specify PSEUDO_QUANTILE, then PROC HPBIN generates a result that approximates the quantile
binning. If you specify WINSOR, PROC HPBIN uses Winsorized binning, and you must specify the
WINSORRATE option with a value from 0.0 to 0.5 exclusive for number . You can specify only one
option. The default is BUCKET.
However, when a BINS_META data set is specified, PROC HPBIN does not do binning and ignores
the binning method options, binning level options, and INPUT statement. Instead, PROC HPBIN
takes the binning results from the BINS_META data set and calculates the weight of evidence and
information value.
COMPUTEQUANTILE
computes the quantile result. If you specify COMPUTEQUANTILE, PROC HPBIN generates the
quantiles and extremes table, which contains the following percentages: 0% (Min), 1%, 5%, 10%, 25%
(Q1), 50% (Median), 75% (Q3), 90%, 95%, 99%, and 100% (Max).
COMPUTESTATS
computes the statistic result. If you specify COMPUTESTATS, basic statistical information is computed
and ODS output can be provided. The output table contains six variables: the mean, median, standard
deviation, minimum, maximum, and number of bins for each binning variable.
DATA=SAS-data-set
specifies the input SAS data set or database table to be used by PROC HPBIN.
If the procedure executes in distributed mode, the input data are distributed to memory on the appliance
nodes and analyzed in parallel, unless the data are already distributed in the appliance database. In this
case, PROC HPBIN reads the data alongside the distributed database.
For single-machine mode, the input must be a SAS data set.
NOPRINT
suppresses the generation of ODS outputs.

NUMBIN=integer
specifies the global number of binning levels for all binning variables. The value of integer can be any
integer between 2 and 1,000, inclusive. The default number of binning levels is 16.
The resulting number of binning levels might be less than the specified integer if the sample size is
small or if the data are not normalized. In this case, PROC HPBIN provides a warning message.
You can specify a different number of binning levels for each different variable in an INPUT statement.
The number of binning levels that you specify in an INPUT statement overwrites the global number of
binning levels.
OUTPUT=SAS-data-set
creates an output SAS data set in single-machine mode or a database table that is saved alongside the
distributed database in distributed mode. The output data set or table contains binning variables. To
avoid data duplication for large data sets, the variables in the input data set are not included in the
output data set.
WOE
computes the weight of evidence (WOE) and information value (IV).
CODE Statement F 49
WOEADJUST=number
specifies the adjustment factor for the weight-of-evidence calculation. You can specify any value from
0.0 to 1.0, inclusive, for number . The default is 0.5.
CODE Statement
CODE FILE=filename ;
The CODE statement is optional in PROC HPBIN. If you use a CODE statement, score code is generated
and stored in a file that can be used for scoring purposes. Only one CODE statement is processed. If you
specify multiple CODE statements, only the first one is used.
FREQ Statement
FREQ variable ;
The variable in the FREQ statement identifies a numeric variable in the data set that contains the frequency
of occurrence for each observation. The HPBIN procedure treats each observation as if it appeared n times,
where n is the value of the variable for the observation. If n is not an integer, it is truncated to an integer. If
n is less than 1 or is missing, the observation is not used in the analysis. When the FREQ statement is not
specified, each observation is assigned a frequency of 1.
ID Statement
ID variables ;
The optional ID statement lists one or more variables from the input data set to be transferred to the output
data set. The ID statement accepts both numeric and character variables. The variables in an ID statement
can also appear in an INPUT statement.
To avoid data duplication for large data sets, the HPBIN procedure does not include any variables from the
input data set in the output data set by default. Therefore, the ID statement can be used to copy variables
from the input data set to the output data set.
INPUT Statement
INPUT variables < / option > ;
The INPUT statement names one or more variables as input variables for binning. The specified variables
must be interval variables. If classification variables are provided, PROC HPBIN stops with an error message.
PROC HPBIN does not support duplicate variables. If the INPUT statement contains a duplicate variable,
PROC HPBIN takes only the first variable and provides a warning message. You can specify the following
option in each INPUT statement:
NUMBIN=integer
specifies the number of binning levels for all the binning variables in the current INPUT statement.
The value of integer can be any integer between 2 and 1,000, inclusive.
The resulting number of binning levels might be less than the specified integer if the sample size is
small or if the data are not normalized. In these cases, PROC HPBIN provides a warning message.
The number of binning levels that you specify in an INPUT statement overwrites the global number of
binning levels. If you do not specify the NUMBIN= option in the INPUT statement, PROC HPBIN
uses the global number of binning levels, which is 16 by default but can be specified in the NUMBIN=
option in the PROC HPBIN statement.
When a BINS_META data set is specified, PROC HPBIN does not do binning and ignores the INPUT
statement. Instead, PROC HPBIN takes the binning results from the BINS_META data set and calculates the
weight of evidence and information value for the interval variables in the BINS_META data set.
the performance characteristics of the HPBIN procedure.
You can also use the PERFORMANCE statement to control whether the HPBIN procedure executes in
single-machine or distributed mode.
The PERFORMANCE statement is documented further in the section PERFORMANCE Statement on
page 36 of Chapter 3, Shared Concepts and Topics.
TARGET Statement
TARGET variable / LEVEL=level ORDER=order ;
The TARGET statement names the variable that PROC HPBIN uses to calculate the weight of evidence and
information value. You can specify the following arguments:
LEVEL=level
specifies the level of the target variable. The values of level can be BINARY or NOMINAL. The
default is NOMINAL.
ORDER=order
specifies the order of the target variable. The values of order can be ASCENDING or DESCENDING.
The default is DESCENDING.
Details: HPBIN Procedure F 51
Details: HPBIN Procedure

Computing the Quantiles (Percentiles)
PROC HPBIN computes the 0% (minimum), 1%, 5%, 10%, 25%, 50%, 75%, 90%, 95%, 99%, and 100%
(maximum) percentiles of each binning variable.
Let m be the number of nonmissing values for a variable, and let x1 , x2 , . . . , xm represent the ordered values
t
of the variable. Let the tth percentile be y, set p D 100
, and let mp = j + g, where j is the integer part of mp
and g is the fractional part of mp. Then the tth percentile is described as:
(
xj
if g D 0
yD
xj C1 if g > 0
N OTE : To avoid the time-consuming sorting process, the HPBIN procedure uses an iterative projection
method to compute percentiles according to the preceding formula.
Binning Computation and Formulas

For variable x , assume that the data set is {xi }, where i D 1; 2; : : : ; n. Let min.x/ D mini 2f1::ng fxi g, and
let max.x/ D maxi 2f1::ng fxi g. The range of the variable is range.x/ D max.x/ min.x/.
The computations for the various binning methods are as follows:
For bucket binning, the length of the bucket is
max.x/
LD
min.x/
n
The split points are

sk D min.x/ C L k
where k D 1; 2; : : : ; numbin
HPBIN statement.
1, and numbin is the value of the NUMBIN= option in the PROC
When the data are evenly distributed on the SAS appliance, the time complexity for bucket binning is
O.n/=.nodes cpus/, where n is the number of observations, nodes is the number of computer nodes
on the appliance, and cpus is the number of CPUs on each node.
For quantile binning, PROC HPBIN calculates a quantile table P . Let P D fpk g, where k D
1; 2; : : : ; numbin. Then pk is described as
8
if k D 0
<1:0=numbin
pk D 1:0=numbin C pk 1 if 0 < k < numbin
:
1:0
if k D numbin
Quantile binning often requires data to be sorted in a particular way, and the sorting process usually
consumes a significant amount of CPU time and memory. When the input data set is larger than the
available memory, the sorting algorithm becomes more complicated. In distributed computing, data
communications overhead also increases the sorting challenge. To avoid the time-consuming sorting
process, the HPBIN procedure uses an iterative projection method for quantile binning, which runs
much faster than sorting-based quantile binning method in most cases.
After calculating the quantile table, PROC HPBIN uses an iterative projection method to compute
quantiles (percentiles) according to the formula that is described in the section Computing the
Quantiles (Percentiles) on page 51.
Quantile binning aims to assign the same number of observations to each bin, if the number of
observations is evenly divisible by the number of bins. As a result, each bin should have the same
number of observations, provided that there are no tied values at the boundaries of the bins. Because
PROC HPBIN always assigns observations that have the same value to the same bin, quantile binning
might create unbalanced bins if any variable has tied values. For example, if an observation whose
value is x is assigned to bin k, then every observation whose value is x is assigned to bin k for this
variable, and no observation whose value is x is assigned to the next bin, bin k C 1. Therefore, bin k
might have more observations than bin k C 1, because the tied values at the boundaries between bin k
and bin k C 1 are all assigned to bin k. That is, tied values at the boundaries between two bins are
always assigned to the lower-numbered bin.
For pseudoquantile binning and Winsorized binning, the sorting algorithm is more complex to describe
but very efficient to execute. For variable x, PROC HPBIN uses a simple bucket sorting method to
obtain the basic information. Let N = 10,000 be the number of buckets, ranging from min.x/ to
max.x/. For each bucket Bi , i D 1; 2; : : : ; N , PROC HPBIN keeps following information:
ci : count of x in Bi
mini : minimum value of x in Bi
maxi : maximum value of x in Bi
P
i : sum of x in Bi
P
2i : sum of x 2 in Bi
To calculate the quantile table, let P D f0:00; 0:01; 0:05; 0:10; 0:25;P
0:50; 0:75; 0:90; 0:95; 0:99; 1:00g.
k
ci pk n. Therefore, the
For each pk 2 P , k D 1; 2; : : : ; 11, find the smallest Ik such that Ii D1
quantile value Qk is obtained,
(
Pk
mi nIk if Ii D1
ci > pk n
Qk D
PIk
maxIk if i D1 ci D pk n
where k D 1; 2; : : : ; 11.
Binning Computation and Formulas F 53
PROC HPBIN calculates the split points as follows:

n
For pseudoquantile binning, let the base count bc D ceil. numbin
/. PROC HPBIN finds those integers
fIk jk D 1; 2; : : : g such that
X
Ik
i D1
ci
IX
k 1
ci C bc
Ik
X
or
i D1
ci
i D1
and
Ik
X
ci >
i D1
and
Ik
X
IX
k 1
n
numbin
k
ci
i D1
ci < n
i D1
where k is the kth split.

The split value is
sk D min.x/ C
max.x/
min.x/
N
Ik
where k D 1; 2; : : : , and k < numbin.

The time complexity for pseudoquantile binning is C O.n/=.nodes cpus/, where C is a constant
that depends on the number of sorting bucket N , n is the number of observations, nodes is the number
of computer nodes on the appliance, and cpus is the number of CPUs on each node.
For Winsorized binning, the Winsorized statistics are computed first. After the minimum and maximum
have been found, the split points are calculated the same way as in bucket binning.
P
Let the tail count wc be ceil.WinsorRate n/, and find the smallest I such that IiD1 ci wc. Then,
P
Pl
the left tail count is lwc D IiD1 ci . Find the next Il such that IiD1
ci > lwc. Therefore, the
P
minimum value is WinsorMin D minIl . Similarly, find the largest I such that N
i DI ci wc. The
PN
PN
right tail count is rwc D i DI ci . Find the next Ir such that i DIr ci > rwc. Then the maximum
value is WinsorMax D maxIr . The mean is calculated by the formula
WinsorMean D
lwc WinsorMin C
PIr
i DIl
sumi C rwc WinsorMax
The trimmed mean is calculated by the formula

PIr
trimmedMean D
i DIl
lwc
sumi
rwc
N OTE : PROC HPBIN prints an error or a warning message when the results might not be accurate.
N OTE : PROC HPBIN does not allow empty bins. If an empty bin is detected because of an insufficient
number of nonmissing observations, PROC HPBIN issues an error and exits.
N OTE : If PROC HPBIN detects an empty bin followed by a bin that is not empty, it skips the empty bin and
does not assign a number to it. In this case, the number of bins that PROC HPBIN generates is less than the
specified number of bins.
Computing the Weight of Evidence and Information Value

PROC HPBIN can compute the weight of evidence and the information value.
Weight of evidence (WOE) is a measure of how much the evidence supports or undermines a hypothesis.
WOE measures the relative risk of an attribute of binning level. The value depends on whether the value of
the target variable is a non-event or an event. An attributes WOE is defined as follows:
WOEat t ri but e D ln
non eve nt
pat
t ri but e
eve nt
pat
t ri but e
D ln
at t ribute
Nnon
event
total
Nnon
event
at t ribute
Nevent
total
Nevent
The definitions of the quantities in the preceding formula are as follows:

at t ri but e : the number of non-event records that exhibit the attribute
Nnon
eve nt
t ot al
Nnon
eve nt : the total number of non-event records
at t ri but e : the number of event records that exhibit the attribute
Neve
nt
t ot al : the total number of event records
Neve
nt
To avoid an undefined WOE, an adjustment factor, x, is used:
WOEat t ri but e D ln
at t ribute
Nnon
event Cx
total
Nnon
event
at t ribute
Nevent
Cx
total
Nevent
You can use the WOEADJUST= option to specify a value between [0, 1] for x. By default, x is 0.5.
The information value (IV) is a weighted sum of the WOE of the characteristics attributes. The weight is the
difference between the conditional probability of an attribute given an event and the conditional probability
of that attribute given a non-event. In the following formula of IV, m is the number of bins of a variable:
IV D
m
but e
X
N at t ri eve
nt
. non
t ot al
N
non eve nt
i D1
at t ri but e
Neve
nt
t ot al
Neve
nt
/ WOEi
An information value can be any real number. Generally speaking, the higher the information value, the more
predictive a characteristic is likely to be.
Data Output
The HPBIN procedure can write an output table back to the database. If the input data are from the appliances
and the processing is alongside the database, PROC HPBIN writes the output table back to each appliance
in parallel. For single-machine mode, the output table is a SAS data set. In this case, you can still take
advantage of multithreading, which performs parallel computation on a single node.
ODS Tables F 55
If you do not specify the OUTPUT option in the PROC HPBIN statement, the write-back process is skipped.
This is useful when only the basic statistical tables are computed.
By default, the output table contains the new binned variables. If you specify an ID statement, the output
table contains the IDs and the binned variables.
N OTE : If an input variable value is missing, then the binning output level value is 0.
ODS Tables
Each table that the HPBIN procedure creates has a name associated with it. You must use this name to refer
to the table in ODS statements. These table names are listed in Table 4.2.
Table 4.2 ODS Tables Produced by PROC HPBIN
Table Name
Description
Statement
Options
BinInfo
PROC HPBIN
Default
InfoValue
Basic binning information and

parameters
Information value for each variable
PROC HPBIN
WOE
Mapping
Level mapping information
PROC HPBIN
Default
NObs
Number of observations read and

used
Information about the
high-performance computing
environment
Quantiles and extremes
PROC HPBIN
WOE
PROC HPBIN
Default
PROC HPBIN
COMPUTEQUANTILE
Summary statistics for the given

variables
Trimmed statistics for the given
variables
PROC HPBIN
COMPUTESTATS
PROC HPBIN
WINSOR,
WINSORRATE
Winsor
Winsor statistics for the given

variables
PROC HPBIN
WINSOR,
WINSORRATE
WOE
Weight of evidence for each bin
PROC HPBIN
WOE
Timing
Timing
PERFORMANCE DETAILS
PerformanceInfo
Quantile
Summary
Trim
The following list provides more information about these tables:

BinInfo
By default, PROC HPBIN generates the Binning Information table in its ODS output.
This table displays some procedure parameters and data information, which includes the
binning method, the number of bins, and the number of variables.
InfoValue
PROC HPBIN generates the Information Value table if you specify the WOE option in
the PROC HPBIN statement. This table provides the information value for each variable.
Mapping
By default, PROC HPBIN generates a Mapping table in its ODS output. This table
provides the level mapping for the input variables. The level starts at 1 and increases to
the value that you specify in the NUMBIN= option. In the mapping table, a missing value
for the lower bound indicates negative infinity, and a missing value for the upper bound
indicates positive infinity.
The final bin level can be less than the NUMBIN value if the input data are small or the
binning variable is discrete. In this case, a warning message is printed in the log.
NObs
PROC HPBIN generates the Nobs table if you specify the WOE option. This table
provides the number of observations that are read and used.
PerformanceInfo
By default, PROC HPBIN produces the Performance Information table. It displays

information about the execution mode. For single-machine mode, the table displays the
number of threads used. For distributed mode, the table displays the grid mode (symmetric
or asymmetric), the number of compute nodes, and the number of threads per node.
Quantile
PROC HPBIN generates the Quantiles and Extremes table if you specify the COMPUTEQUANTILE option. This table contains the following quantile levels for each
variable: 0% (Min), 1%, 5%, 10%, 15%, 25% (Q1), 50% (Median), 75% (Q3), 90%, 95%,
99%, and 100% (Max).
Summary
PROC HPBIN generates the Summary Statistics table if you specify the COMPUTESTATS option. This table displays the variable name, number of nonmissing observations,
number of missing observations, mean, median, standard deviation, minimum, maximum,
and number of bins.
Trim
PROC HPBIN also generates the Trimmed Statistics table if you specify the WINSOR
option. This table contains the trimmed minimum, maximum, mean, standard error mean,
left tail, left tail percentage, right tail, and right tail percentage, and the degrees of freedom
(DF).
Winsor
PROC HPBIN generates the Winsor Statistics table if you specify the WINSOR option.
The Winsor Statistics table contains the Winsorized minimum, maximum, mean, standard error mean, left tail, left tail percentage, right tail, and right tail percentage, and the
degrees of freedom (DF).
WOE
PROC HPBIN generates the Weight of Evidence table if you specify the WOE option in
the PROC HPBIN statement. This table provides the level mapping information, binning
information, weight of evidence, and information value for each bin.
When the Weight of Evidence table is printed, the Mapping table is not printed
because the level mapping information is the same in both tables.
In addition to the level mapping information, the Weight of Evidence table contains
some other information such as the non-event count, non-event rate, event count, event
rate, weight of evidence, and information value for each bin.
Example 4.1: Bucket Binning in Single-Machine Mode F 57
Examples
Example 4.1: Bucket Binning in Single-Machine Mode
This example shows how you can use PROC HPBIN in single-machine mode.
The following DATA step creates the SAS data set as the input for the example:
data ex12;
length id 8;
do id=1 to 1000;
x1 = ranuni(101);
x2 = 10*ranuni(201);
x3 = 100*ranuni(301);
output;
end;
run;
The following statements show the basic usage:

proc hpbin data=ex12 output=out numbin=10 bucket ;
input x1-x3;
run;
proc print data=out(obs=10); run;
PROC HPBIN generates several ODS tables, which are shown in Output 4.1.1 through Output 4.1.2.
Output 4.1.1 PROC HPBIN Binning Information Display
The HPBIN Procedure

Binning Information
Method
Bucket Binning
Number of Bins Specified 10

Number of Variables
Output 4.1.2 PROC HPBIN Mapping Table

Mapping
Binned
Variable Variable Range
x1
BIN_x1 x1 < 0.099925
x3
108 0.10800000
0.099925 <= x1 < 0.199399
94 0.09400000
0.199399 <= x1 < 0.298872
90 0.09000000
0.298872 <= x1 < 0.398345
95 0.09500000
0.398345 <= x1 < 0.497818
90 0.09000000
0.497818 <= x1 < 0.597291
115 0.11500000
0.597291 <= x1 < 0.696764
116 0.11600000
0.696764 <= x1 < 0.796237
105 0.10500000
0.796237 <= x1 < 0.895711
106 0.10600000
0.895711 <= x1
x2
Frequency Proportion
BIN_x2 x2 < 1.021539
81 0.08100000
99 0.09900000
1.021539 <= x2 < 2.015407
110 0.11000000
2.015407 <= x2 < 3.009275
99 0.09900000
3.009275 <= x2 < 4.003143
85 0.08500000
4.003143 <= x2 < 4.997011
109 0.10900000
4.997011 <= x2 < 5.990878
98 0.09800000
5.990878 <= x2 < 6.984746
100 0.10000000
6.984746 <= x2 < 7.978614
94 0.09400000
7.978614 <= x2 < 8.972482
112 0.11200000
8.972482 <= x2
94 0.09400000
BIN_x3 x3 < 10.173597
97 0.09700000
10.173597 <= x3 < 20.142996
97 0.09700000
20.142996 <= x3 < 30.112396
106 0.10600000
30.112396 <= x3 < 40.081796
94 0.09400000
40.081796 <= x3 < 50.051196
102 0.10200000
50.051196 <= x3 < 60.020596
99 0.09900000
60.020596 <= x3 < 69.989995
107 0.10700000
69.989995 <= x3 < 79.959395
92 0.09200000
79.959395 <= x3 < 89.928795

89.928795 <= x3
91 0.09100000
115 0.11500000
In this example, PROC HPBIN also generates the output table. The first 10 observations of the table are
shown in Output 4.1.3.
Example 4.2: PseudoQuantile Binning in Distributed Mode F 59
Output 4.1.3 First 10 Observations in the Output Table

Obs BIN_x1 BIN_x2 BIN_x3
1
10
10
10
Example 4.2: PseudoQuantile Binning in Distributed Mode

This example shows pseudoquantile binning that is executed in distributed mode. The following DATA step
generates 1,000,000 observations:
data ex12;
length id 8;
do id=1 to 1000000;
x1 = ranuni(101);
x2 = 10*ranuni(201);
output;
end;
run;
You can run PROC HPBIN in distributed mode by specifying valid values for the NODES=, INSTALL=,
and HOST= options in the PERFORMANCE statement. An alternative to specifying the INSTALL= and
HOST= options in the PERFORMANCE statement is to set appropriate values for the GRIDHOST and
GRIDINSTALLLOC environment variables by using OPTIONS SET commands. See the section Processing
Modes on page 10 in Chapter 3, Shared Concepts and Topics, for details about setting these options or
environment variables.
The following statements provide an example. To run these statements successfully, you need to set the
macro variables GRIDHOST and GRIDINSTALLLOC to resolve to appropriate values, or you can replace the
references to macro variables with appropriate values.
ods output BinInfo=bininfo;

ods output PerformanceInfo=perfInfo;
ods output Mapping=mapTable;
ods output Summary=Summary;
ods output Quantile=Quantile;
ods listing close;
proc hpbin data=ex12 output=out numbin=10 pseudo_quantile
computestats computequantile ;
input x1-x2;
performance nodes=4 nthreads=8
host="&GRIDHOST" install="&GRIDINSTALLLOC";
run;
ods listing;
The Performance Information table in Output 4.2.1 shows the grid setting.
Output 4.2.1 PROC HPBIN Performance Information
Host Node
<< your grid host >>
Install Location
<< your grid install location >>
Execution Mode
Distributed
Grid Mode
Symmetric
The Binning Information table in Output 4.2.2 shows the binning method, number of bins, and number of
variables.
Output 4.2.2 PROC HPBIN Binning Information
Binning Information
Method
Pseudo-Quantile Binning

Number of Variables
The Mapping table in Output 4.2.3 shows the level mapping of the input variables.
Example 4.2: PseudoQuantile Binning in Distributed Mode F 61
Output 4.2.3 PROC HPBIN Mapping

Mapping
Binned
x1
BIN_x1 x1 < 0.099900
100046 0.10004600
0.099900 <= x1 < 0.199500
100029 0.10002900
0.199500 <= x1 < 0.299300
100016 0.10001600
0.299300 <= x1 < 0.399500
99939 0.09993900
0.399500 <= x1 < 0.500000
100049 0.10004900
0.500000 <= x1 < 0.599800
99989 0.09998900
0.599800 <= x1 < 0.700400
99975 0.09997500
0.700400 <= x1 < 0.800300
100014 0.10001400
0.800300 <= x1 < 0.900299
100007 0.10000700
0.900299 <= x1
x2
99936 0.09993600
BIN_x2 x2 < 0.997008
100006 0.10000600
0.997008 <= x2 < 1.995006
100025 0.10002500
1.995006 <= x2 < 2.994005
99986 0.09998600
2.994005 <= x2 < 3.995004
100034 0.10003400
3.995004 <= x2 < 4.999002
99990 0.09999000
4.999002 <= x2 < 5.998001
100063 0.10006300
5.998001 <= x2 < 6.993000
99929 0.09992900
6.993000 <= x2 < 7.998998
100008 0.10000800
7.998998 <= x2 < 8.999997
100010 0.10001000
8.999997 <= x2
99949 0.09994900
The Summary Statistics table in Output 4.2.4 displays the basic statistical information, including the
number of observations, number of missing observations, mean, median, and so on.
Output 4.2.4 PROC HPBIN Summary Statistics Table
Summary Statistics
Variable
N N Missing
Mean
Median
Std Dev
Minimum
Maximum N Bins
x1
1000000
0 0.49984213 0.49991238 0.28894736 2.24449E-7 0.99999939
10
x2
1000000
0 4.99688234 4.99851593 2.88736227 9.10833E-6 9.99999537
10
The Quantiles and Extremes table in Output 4.2.5 shows the quantile computation of the variables. The
ODS table is generated only when the COMPUTESTATS option is specified in the PROC HPBIN statement.
Output 4.2.5 PROC HPBIN Quantile Computation

Quantiles and Extremes
Variable Quantile Level
x1
x2
Quantile
Max
0.99999939
.99
0.99011639
.95
0.95024946
.90
0.90023557
.75 Q3
0.75032495
.50 Median
0.49991238
.25 Q1
0.24931534
.10
0.09985729
.05
0.04954403
.01
0.01000524
Min
2.24449E-7
Max
9.99999537
.99
9.90136979
.95
9.49989152
.90
8.99939011
.75 Q3
7.49894200
.50 Median
4.99851593
.25 Q1
2.49431827
.10
0.99691767
.05
0.49879104
.01
0.10062442
Min
9.10833E-6
Example 4.3: Quantile Binning in Distributed Mode

This example shows quantile binning that is executed in distributed mode. Most of this example is the same
as the pseudoquantile binning example (see Example 4.2: PseudoQuantile Binning in Distributed Mode),
so you can easily compare these two binning methods. The following DATA step generates 1,000,000
observations:
data ex12;
length id 8;
do id=1 to 1000000;
x1 = ranuni(101);
x2 = 10*ranuni(201);
output;
end;
run;
You can run PROC HPBIN in distributed mode by specifying valid values for the NODES=, INSTALL=,
and HOST= options in the PERFORMANCE statement. An alternative to specifying the INSTALL= and
HOST= options in the PERFORMANCE statement is to set appropriate values for the GRIDHOST and
Example 4.3: Quantile Binning in Distributed Mode F 63
GRIDINSTALLLOC environment variables by using OPTIONS SET commands. See the section Processing
Modes on page 10 in Chapter 3, Shared Concepts and Topics, for details about setting these options or
environment variables.
ods output BinInfo=bininfo;
ods output PerformanceInfo=perfInfo;
ods output Mapping=mapTable;
ods listing close;
proc hpbin data=ex12 output=out numbin=10 quantile;
input x1-x2;
performance nodes=4 nthreads=8
run;
ods listing;
The Performance Information table in Output 4.3.1 shows the grid setting.
Output 4.3.1 PROC HPBIN Performance Information
Host Node
Install Location
Execution Mode
Distributed
Grid Mode
Symmetric
The Binning Information table in Output 4.3.2 shows the binning method, number of bins, and number of
variables.
Output 4.3.2 PROC HPBIN Binning Information
Binning Information
Method
Quantile Binning

Number of Variables
The Mapping table in Output 4.3.3 shows the level mapping of the input variables. As you can see from
this table, when the binning method is quantile, PROC HPBIN assigns the same number of observations to
each bin for the input variables if possible.
Output 4.3.3 PROC HPBIN Mapping

Mapping
Binned
x1
x2
BIN_x1 x1 < 0.099859
100000 0.10000000
0.099859 <= x1 < 0.199413
100000 0.10000000
0.199413 <= x1 < 0.299210
100000 0.10000000
0.299210 <= x1 < 0.399472
100000 0.10000000
0.399472 <= x1 < 0.499913
100000 0.10000000
0.499913 <= x1 < 0.599746
100000 0.10000000
0.599746 <= x1 < 0.700361
100000 0.10000000
0.700361 <= x1 < 0.800231
100000 0.10000000
0.800231 <= x1 < 0.900236
100000 0.10000000
0.900236 <= x1
100000 0.10000000
BIN_x2 x2 < 0.996924
100000 0.10000000
0.996924 <= x2 < 1.994716
100000 0.10000000
1.994716 <= x2 < 2.993747
100000 0.10000000
2.993747 <= x2 < 3.994634
100000 0.10000000
3.994634 <= x2 < 4.998520
100000 0.10000000
4.998520 <= x2 < 5.997022
100000 0.10000000
5.997022 <= x2 < 6.992673
100000 0.10000000
6.992673 <= x2 < 7.998557
100000 0.10000000
7.998557 <= x2 < 8.999391
100000 0.10000000
8.999391 <= x2
100000 0.10000000
Example 4.4: Winsorized Binning

For Winsorized binning, PROC HPBIN provides bucket binning and basic Winsorized statistical information
for the input data. The following statements include the WINSOR and WINSORRATE=0.05 options and
generate tables for Winsor and trimmed statistics.
data ex12;
length id 8;
do id=1 to 10000;
x1 = ranuni(101);
x2 = 10*ranuni(201);
x3 = 100*ranuni(301);
output;
end;
run;
ods output Winsor=Winsor;
ods output Trim=Trim;
ods listing close;
proc hpbin data=ex12 NUMBIN=10 WINSOR WINSORRATE=0.05;
input x1-x2;
run;
ods listing;
The preceding statements generate Output 4.4.1 and Output 4.4.2.
Example 4.5: Bucket Binning and Weight-of-Evidence Computation F 65
Output 4.4.1 PROC HPBIN Winsorized Statistics
The HPBIN Procedure

Winsorized Statistics
Variable
Mean
Std Error
N Percent
N
Mean Left Tail Left Tail Right Tail
Percent
Right Tail
DF
x1
0.50175743 0.00316436
500
501 5.01000000 8998
x2
5.03025502 0.03185024
500
501 5.01000000 8998
Output 4.4.2 PROC HPBIN Trimmed Statistics

Trimmed Statistics
Variable
Mean
Std Error
N Percent
N
Mean Left Tail Left Tail Right Tail
Percent
Right Tail
DF
x1
0.50209276 0.00316434
500
501 5.01000000 8998
x2
5.03201037 0.03185006
500
501 5.01000000 8998
Example 4.5: Bucket Binning and Weight-of-Evidence Computation

This example shows bucket binning and weight-of-evidence (WOE) computation in two steps. In the first
step, PROC HPBIN does bucket binning and generates the mapping table. In the second step, PROC HPBIN
takes the mapping table that is generated in the first step as the BINS_META data set and computes the WOE
and the information value (IV).
data ex12;
input cl1 $ x0
cards;
a
2
a
2
a
3
c
2
c
2
a
3
b
1
b
2
b
1
b
2
run;
x1
x2
.
2
0
3
.
6
4
5
6
3
y $ freq id;
7
6
1
7
4
7
4
6
4
2
n
.
o
y
n
n
y
y
o
n
2
3
0
.
-5
3
4
3
1
3
1
2
3
4
5
6
7
8
9
10
proc hpbin data=ex12 numbin=5;

input x1/numbin=4;
input x2;
ods output Mapping=Mapping;
run;
proc hpbin data=ex12 WOE BINS_META=Mapping;
target y/level=nominal order=desc;
run;
Output 4.5.1 Number of Observations Table
The HPBIN Procedure

Number of Observations Read 10
Number of Observations Used
Output 4.5.2 Weight-of-Evidence Table

Weight of Evidence
Binned
x1
x2
Non-event Non-event Event

Count
Rate Count
BIN_x1
Event
Rate
Weight of Information
Evidence
Value
0 0.91629073 0.30543024
x1 < 1.500000
0 0.40546511 0.06757752
1.500000 <= x1 < 3.000000
3.000000 <= x1 < 4.500000
1 0.33333333
2 0.66666667 -1.3862944 0.69314718
4.500000 <= x1
2 0.66666667
1 0.33333333
x2 < 2.200000
0 0.91629073 0.30543024
2.200000 <= x2 < 3.400000
3.400000 <= x2 < 4.600000
2 0.66666667
1 0.33333333
4.600000 <= x2 < 5.800000
5.800000 <= x2
2 0.50000000
BIN_x2
Variable
Information
Value
x1
1.06615494
x2
0.53647930
2 0.50000000 -0.6931472 0.23104906
Output 4.5.3 Variable Information Value Table

Variable Information
Value
Chapter 5
The HPCORR Procedure

Contents
Overview: HPCORR Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
PROC HPCORR Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
Getting Started: HPCORR Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
Syntax: HPCORR Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
PROC HPCORR Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
71
FREQ Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
73
VAR Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
WEIGHT Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
WITH Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
Details: HPCORR Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
Pearson Product-Moment Correlation . . . . . . . . . . . . . . . . . . . . . . . . . .
74
Multithreading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
Output Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
Output Data Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
77
ODS Table Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
Examples: HPCORR Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
Example 5.1: Computing the Pearson Measure of Association in Single-Machine Mode
78
Example 5.2: Computing the Pearson Measure of Association in Distributed Mode . .
80
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
Overview: HPCORR Procedure

The HPCORR procedure computes Pearson correlation coefficients and the probabilities associated with
these statistics. The Pearson product-moment correlation is a parametric measure of a linear relationship
between two variables.
A related type of correlation, polychoric correlation, measures the correlation between two unobserved
variables that have a bivariate normal distribution. Information about these variables is obtained through two
corresponding observed ordinal variables that are derived from the unobserved variables by classifying their
values into finite sets of discrete, ordered values. Polychoric correlation is not available in the HPCORR
procedure, but it is available in the FREQ procedure.
When only one set of analysis variables is specified, the default correlation analysis includes descriptive
statistics for each analysis variable and pairwise Pearson correlation statistics for these variables.
68 F Chapter 5: The HPCORR Procedure
When two sets of analysis variables are specified, the default correlation analysis includes descriptive statistics
for each analysis variable and pairwise Pearson correlation statistics between the two sets of variables.
When the relationship between two variables is nonlinear or when outliers are present, the correlation
coefficient might incorrectly estimate the strength of the relationship.
You can save the correlation statistics in a SAS data set for use with other statistical and reporting procedures.
PROC HPCORR runs in either single-machine mode or distributed mode.
PROC HPCORR Features

The HPCORR procedure is a high-performance procedure that offers the following functionality:
It can perform analysis on a massively parallel SAS high-performance appliance.
It reads input data in parallel and writes output data in parallel when the data source is the appliance
database.
It is highly multithreaded during all phases of analytic execution.
Because the HPCORR procedure is a high-performance analytical procedure, it also does the following:
computations
Topics.
Getting Started: HPCORR Procedure

This example creates a simple data set and then uses PROC HPCORR to produce simple Pearson correlations
by executing on the client machine.
The following statements create the data set Fitness, which has been altered to contain some missing values:
*----------------- Data on Physical Fitness -----------------*
| These measurements were made on men involved in a physical |
| fitness course at N.C. State University.
|
| The variables are Age (years), Weight (kg),
|
| Runtime (time to run 1.5 miles in minutes), and
|
| Oxygen (oxygen intake, ml per kg body weight per minute)
|
| Certain values were changed to missing for the analysis.
|
-----------------------------------------------------------*
*;
Getting Started: HPCORR Procedure F 69
data Fitness;
input Age Weight Oxygen RunTime @@;
datalines;
44 89.47 44.609 11.37
40 75.07 45.313
44 85.84 54.297 8.65
42 68.15 59.571
38 89.02 49.874
.
47 77.45 44.811
40 75.98 45.681 11.95
43 81.19 49.091
44 81.42 39.442 13.08
38 81.87 60.055
44 73.03 50.541 10.13
45 87.66 37.388
45 66.45 44.754 11.12
47 79.15 47.273
54 83.12 51.855 10.33
49 81.42 49.156
51 69.63 40.836 10.95
51 77.91 46.672
48 91.63 46.774 10.25
49 73.37
.
57 73.37 39.407 12.63
54 79.38 46.080
52 76.32 45.441 9.63
50 70.87 54.625
51 67.25 45.118 11.08
54 91.63 39.203
51 73.71 45.790 10.47
57 59.08 50.545
49 76.32
.
.
48 61.24 47.920
52 82.78 47.467 10.50
;
10.07
8.17
11.63
10.85
8.63
14.03
10.60
8.95
10.00
10.08
11.17
8.92
12.88
9.93
11.50
The following statements invoke the HPCORR procedure and request a correlation analysis:
proc hpcorr data=Fitness;
run;
The Performance Information table in Figure 5.1 shows that the procedure executes in single-machine
modethat is, the data reside and the computation executes on the machine where the SAS session executes.
This run of the HPCORR procedure was performed on a multicore machine; one computational thread was
spawned for each core.
The Simple Statistics table in Figure 5.1 displays univariate statistics for the analysis variables.
Figure 5.1 Performance Information and Univariate Statistics

Execution Mode
Single-Machine
Number of Threads 4
4 Variables: Age
Weight Oxygen RunTime
Simple Statistics
Variable
Mean Std Dev
Sum Minimum Maximum
Age
31 47.67742 5.21144
1478 38.00000 57.00000
Weight
31 77.44452 8.32857
2401 59.08000 91.63000
Oxygen
29 47.22721 5.47718
1370 37.38800 60.05500
RunTime 29 10.67414 1.39194 309.55000
8.17000 14.03000
By default, all numeric variables not listed in other statements are used in the analysis. Observations that
have nonmissing values for each variable are used to derive the univariate statistics for that variable.
The Pearson Correlation Coefficients table in Figure 5.2 displays the Pearson correlation, the p-value under
the null hypothesis of zero correlation, and the number of nonmissing observations for each pair of variables.
Figure 5.2 Pearson Correlation Coefficients

Pearson Correlation Coefficients
Prob > |r| under H0: Rho=0
Number of Observations
Age Weight Oxygen RunTime
Age
1.00000 -0.23354 -0.31474

0.2061 0.0963
31
31
29
0.14478
0.4536
29
Weight
-0.23354 1.00000 -0.15358

0.2061
0.4264
31
31
29
0.20072
0.2965
29
Oxygen
-0.31474 -0.15358 1.00000 -0.86843

0.0963 0.4264
<.0001
29
29
29
28
RunTime 0.14478 0.20072 -0.86843

0.4536 0.2965 <.0001
29
29
28
1.00000
29
By default, Pearson correlation statistics are computed from observations that have nonmissing values for
each pair of analysis variables. Figure 5.2 displays a correlation of 0.86843 between Runtime and Oxygen,
which is significant with a p-value less than 0.0001. That is, an inverse linear relationship exists between
these two variables. As Runtime (time in minutes to run 1.5 miles) increases, Oxygen (oxygen intake in
milliliters per kilogram body weight per minute) decreases.
Syntax: HPCORR Procedure

The following statements are available in PROC HPCORR:
PROC HPCORR < options > ;
FREQ variable ;
PERFORMANCE performance-options ;
VAR variables ;
WEIGHT variable ;
WITH variables ;
The FREQ statement specifies the variable that represents the frequency of occurrence for other values in the
observation.
The VAR statement lists the numeric variables to be analyzed and their order in the correlation matrix. If you
omit the VAR statement, all numeric variables not listed in other statements are used.
The WEIGHT statement identifies the variable whose values weight each observation to compute Pearson
product-moment correlation.
The WITH statement lists the numeric variables with which correlations are to be computed.
The PROC HPCORR statement is the only required statement for the HPCORR procedure.
The rest of this section provides detailed syntax information for each of these statements, beginning with the
PROC HPCORR statement. The remaining statements are presented in alphabetical order.
PROC HPCORR Statement F 71
PROC HPCORR Statement

PROC HPCORR < options > ;
Table 5.1 summarizes the options available in the PROC HPCORR statement.
Table 5.1 Summary of PROC HPCORR Options
Option
Description
Data Sets
DATA=
OUTP=
Specifies the output data set with Pearson correlation statistics
Statistical Analysis
EXCLNPWGT Excludes from the analysis observations that have nonpositive weight values
NOMISS
Excludes from the analysis observations that have missing analysis values
Pearson Correlation Statistics
COV
Computes covariances
CSSCP
Computes corrected sums of squares and crossproducts
SSCP
Computes sums of squares and crossproducts
Printed Output
BEST=
Displays the specified number of ordered correlation coefficients
NOCORR
Suppresses Pearson correlations
NOPRINT
Suppresses all printed output
NOPROB
Suppresses p-values
NOSIMPLE
Suppresses descriptive statistics
RANK
Displays ordered correlation coefficients
You can specify the following options in the PROC HPCORR statement:
BEST=n
prints the n highest correlation coefficients for each variable. The value of n must be greater than or
equal to 1. Correlations are ordered from highest to lowest in absolute value. If you do not specify this
option, PROC HPCORR prints correlations in a rectangular table, using the variable names as row and
column labels.
COV
displays the variance/covariance matrix. When you specify this option, the Pearson correlations are
also displayed. If you specify the OUTP= option, the output data set also contains the covariance
matrix with the corresponding _TYPE_ variable value 'COV'.
CSSCP
displays a table of the corrected sums of squares and crossproducts. When you specify this option,
the Pearson correlations are also displayed. If you specify the OUTP= option, the output data set also
contains a CSSCP matrix with the corresponding _TYPE_ variable value 'CSSCP'.
DATA=SAS-data-set
names the SAS data set to be analyzed by PROC HPCORR. If you do not specify this option, PROC
HPCORR uses the most recently created SAS data set.
EXCLNPWGT
EXCLNPWGTS
excludes from the analysis observations that have nonpositive weight values. If you do not specify this
option, PROC HPCORR treats observations with negative weights like those with zero weights and
counts them in the total number of observations.
NOCORR
suppresses the display of the Pearson correlations. If you specify the OUTP= option, the data set
_TYPE_ remains 'CORR'. To change the data set type to COV, CSSCP, or SSCP, use the (TYPE=)
data set option.
NOMISS
excludes from the analysis observations that have missing values. If you do not specify this option,
PROC HPCORR computes correlation statistics by using all of the nonmissing pairs of variables.
Using this option is computationally more efficient.
NOPRINT
suppresses all displayed output. Use this option if you only want to create an output data set.
NOPROB
suppresses the display of the probabilities that are associated with each correlation coefficient.
NOSIMPLE
suppresses the printing of simple descriptive statistics for each variable. However, if you request an
output data set, the output data set still contains simple descriptive statistics for the variables.
OUTP=output-data-set
OUT=output-data-set
creates an output data set that contains Pearson correlation statistics. This data set also includes means,
standard deviations, and the number of observations. The value of the _TYPE_ variable is 'CORR'.
RANK
displays the ordered correlation coefficients for each variable. Correlations are ordered from highest to
lowest in absolute value.
SSCP
displays a table of the sums of squares and crossproducts. When you specify this option, the Pearson
correlations are also displayed. If you specify the OUTP= option, the output data set contains a SSCP
matrix and the corresponding _TYPE_ variable value is 'SSCP'.
VARDEF=DF | N | WDF | WEIGHT | WGT
specifies the variance divisor in the calculation of variances and covariances. The default is
VARDEF=DF.
Table 5.2 shows the available values and associated divisors for the VARDEF= option, where n is
the number of nonmissing observations and wj is the weight associated with the jth nonmissing
observation.
FREQ Statement F 73
Table 5.2 Possible Values for the VARDEF= Option
Value
Description
Divisor
DF
N
WDF
WEIGHT | WGT
Degrees of freedom
Number of observations
Sum of weights minus one
Sum of weights
n 1
n
Pn
w
Pnj j
w
j j
FREQ Statement
FREQ variable ;
The FREQ statement specifies a numeric variable whose value represents the frequency of the observation.
If you use the FREQ statement, PROC HPCORR assumes that each observation represents n observations,
where n is the value of the FREQ variable. If n is not an integer, SAS truncates it. If n is less than 1 or is
missing, the observation is excluded from the analysis. The sum of the frequency variables represents the
total number of observations.
The effects of the FREQ and WEIGHT statements are similar except when the degrees of freedom are
calculated.
the performance characteristics of the HPCORR procedure.
You can also use the PERFORMANCE statement to control whether the HPCORR procedure executes in
single-machine mode or distributed mode.
page 36.
VAR Statement
VAR variables ;
The VAR statement lists variables for which correlation coefficients are to be computed. If the VAR statement
is not specified, PROC HPCORR computes correlations for all numeric variables that are not listed in other
statements.
WEIGHT Statement
WEIGHT variable ;
The WEIGHT statement lists weights to use in the calculation of Pearson weighted product-moment correlation.
Observations that have missing weights are excluded from the analysis. By default, for observations that
have nonpositive weights, weights are set to 0 and the observations are included in the analysis. You can use
the EXCLNPWGT option to exclude observations with negative or zero weights from the analysis.
WITH Statement
WITH variables ;
The WITH statement lists variables with which correlations of the VAR statement variables are to be
computed. The WITH statement requests correlations of the form r.Xi ; Yj /, where X1 ; : : : ; Xm are analysis
variables that you specify in the VAR statement and Y1 ; : : : ; Yn are variables that you specify in the WITH
statement. The correlation matrix has a rectangular structure of the form
3
2
r.Y1 ; X1 / r.Y1 ; Xm /
7
6
::
::
::
4
5
:
:
:
r.Yn ; X1 / r.Yn ; Xm /
For example, the statements
proc corr;
var x1 x2;
with y1 y2 y3;
run;
produce correlations for the following combinations:

2
3
r.Y 1; X1/ r.Y 1; X 2/
4 r.Y 2; X1/ r.Y 2; X 2/ 5
r.Y 3; X1/ r.Y 3; X 2/
Details: HPCORR Procedure

Pearson Product-Moment Correlation
The Pearson product-moment correlation is a parametric measure of association for two variables. It measures
both the strength and the direction of a linear relationship. If one variable X is an exact linear function of
another variable Y, a positive relationship exists if the correlation is 1, and a negative relationship exists if the
correlation is 1. If there is no linear predictability between the two variables, the correlation is 0. If the
Pearson Product-Moment Correlation F 75
two variables are normal with a correlation 0, the two variables are independent. Correlation does not imply
causality because, in some cases, an underlying causal relationship might not exist.
The formula for the population Pearson product-moment correlation, denoted xy , is
E. .x E.x//.y E.y// /
Cov.x; y/
xy D p
Dp
V.x/V.y/
E.x E.x//2 E.y E.y//2
The sample correlation, such as a Pearson product-moment correlation or weighted product-moment correlation, estimates the population correlation. The formula for the sample Pearson product-moment correlation is
as follows, where xN is the sample mean of x is the sample mean of y:
P
. .xi x/.y
N i y/
N /
rxy D pP i
P
N 2
x/
N 2 i .yi y/
i .xi
The formula for a weighted Pearson product-moment correlation is as follows, where wi is the weight, xN w is
the weighted mean of x, and yNw is the weighted mean of y:
P
wi .xi xN w /.yi yNw /
rxy D pP i
P
xN w /2 i wi .yi yNw /2
i wi .xi
Probability Values
Probability values for the Pearson correlation are computed by treating the following equation as if it came
from a t distribution with .n 2/ degrees of freedom, where r is the sample correlation:
t D .n
2/
1=2
r2
1 r2
1=2
The partial variance-covariance matrix is calculated with the variance divisor (specified in the VARDEF=
option). PROC HPCORR then uses the standard Pearson correlation formula on the partial variancecovariance matrix to calculate the Pearson partial correlation matrix.
When a correlation matrix is positive definite, the resulting partial correlation between variables x and y after
adjusting for a single variable z is identical to that obtained from the following first-order partial correlation
formula, where rxy , rxz , and ryz are the appropriate correlations:
rxy:z D q
rxy
.1
rxz ryz
2 /.1
rxz
2 /
ryz
The formula for higher-order partial correlations is a straightforward extension of the preceding first-order
formula. For example, when the correlation matrix is positive definite, the partial correlation between x and
y that controls for both z_1 and z_2 is identical to the following second-order partial correlation formula,
where rxy:z1 , rxz2 :z1 , and ryz2 :z1 are first-order partial correlations among variables x, y, and z_2 given z_1:
rxy:z1 z2 D q
rxy:z1
.1
rxz2 :z1 ryz2 :z1
2
rxz
/.1
2 :z1
2
ryz
/
2 :z1
Multithreading
Threading refers to the organization of computational work into multiple tasks (processing units that can
be scheduled by the operating system). A task is associated with a thread. Multithreading refers to the
concurrent execution of threads. When multithreading is possible, substantial performance gains can be
realized compared to sequential (single-threaded) execution.
The number of threads spawned by the HPCORR procedure is determined by the number of CPUs on a
machine and can be controlled in the following ways:
You can specify the CPU count with the CPUCOUNT= SAS system option. For example, if you
specify the following statements, the HPCORR procedure schedules threads as if it executed on a
system with four CPUs, regardless of the actual CPU count:
options cpucount=4;
You can specify the NTHREADS= option in the PERFORMANCE statement to determine the number
of threads. This specification overrides the NOTHREADS system option. Specify NTHREADS=1 to
force single-threaded execution.
The number of threads per machine is displayed in the Performance Information table, which is part of the
default output. The HPCORR procedure allocates one thread per CPU.
The HPCORR procedure implements a data-parallel model. For example, if the input data set has 1,000
observations and you are running with four threads, then 250 observations are associated with each thread.
All operations that require access to the data are then multithreaded.
Output Tables
By default, PROC HPCORR prints a report that includes descriptive statistics and correlation statistics for
each variable. The descriptive statistics include the number of observations that have nonmissing values, the
mean, the standard deviation, the minimum, and the maximum.
If a nonparametric measure of association is requested, the descriptive statistics include the median. Otherwise,
the sample sum is included. If a Pearson partial correlation is requested, the descriptive statistics also include
the partial variance and partial standard deviation.
If variable labels are available, PROC HPCORR labels the variables. If you specify the CSSCP, SSCP, or
COV option, the appropriate sums of squares and crossproducts and covariance matrix appear at the top of
the correlation report. If the data set contains missing values, PROC HPCORR prints additional statistics for
each pair of variables.
These statistics, calculated from the observations that have nonmissing row and column variable values,
might include the following:
SSCP(W,V), uncorrected sums of squares and crossproducts
Output Data Sets F 77
USS(W), uncorrected sums of squares for the row variable

USS(V), uncorrected sums of squares for the column variable
CSSCP(W,V), corrected sums of squares and crossproducts
CSS(W), corrected sums of squares for the row variable
CSS(V), corrected sums of squares for the column variable
COV(W,V), covariance
VAR(W), variance for the row variable
VAR(V), variance for the column variable
DF(W,V), divisor for calculating covariance and variances
For each pair of variables, PROC HPCORR prints the correlation coefficients, the number of observations
that are used to calculate the coefficient, and the p-value.
Output Data Sets

PROC HPCORR creates an output data set that contains statistics for Pearson correlation. By default, the
output data set is a special data set type (TYPE=CORR) that is recognized by many SAS/STAT procedures,
including the REG and FACTOR procedures. When you specify the NOCORR option and the COV, CSSCP,
or SSCP option, use the (TYPE=) data set option to change the data set type to COV, CSSCP, or SSCP.
The output data set includes the following variables:
BY variables, which identify the BY group when a BY statement is used
_TYPE_ variable, which identifies the type of observation
_NAME_ variable, which identifies the variable that corresponds to a given row of the correlation
matrix
INTERCEPT variable, which identifies variable sums when the SSCP option is specified
VAR variables, which identify the variables listed in the VAR statement
You can use a combination of the _TYPE_ and _NAME_ variables to identify the contents of an observation.
The _NAME_ variable indicates which row of the correlation matrix the observation corresponds to. The
values of the _TYPE_ variable are as follows:
SSCP, uncorrected sums of squares and crossproducts
CSSCP, corrected sums of squares and crossproducts
COV, covariances
MEAN, mean of each variable

STD, standard deviation of each variable
N, number of nonmissing observations for each variable
SUMWGT, sum of the weights for each variable when using a WEIGHT statement
HPCORR, correlation statistics for each variable
If you specify the SSCP option, the OUTP= data set includes an additional observation that contains intercept
values.
ODS Table Names

PROC HPCORR assigns a name to each table it creates. These names are listed in Table 5.3. You must use
these names to refer to tables when you use the Output Delivery System (ODS). For more information about
ODS, see Chapter 20, Using the Output Delivery System (SAS/STAT Users Guide).
Table 5.3 ODS Tables Produced by PROC HPCORR
ODS Table Name
Description
Option
Cov
Csscp
PearsonCorr
SimpleStats
Sscp
VarInformation
PerformanceInfo
Covariances
Corrected sums of squares and crossproducts
Pearson correlations
Simple descriptive statistics
Sums of squares and crossproducts
Variable information
Information about high-performance computing environment
COV
CSSCP
SSCP
Examples: HPCORR Procedure

Example 5.1: Computing the Pearson Measure of Association in
Single-Machine Mode
The Fitness data set created in the section Getting Started: HPCORR Procedure on page 68 contains
measurements from a study of physical fitness of 31 participants. The following statements request the
Pearson measure of association for the variables Weight, Oxygen, and Runtime:
title 'Measures of Association for a Physical Fitness Study';
proc hpcorr data=Fitness pearson;
var Weight Oxygen RunTime;
run;
Example 5.1: Computing the Pearson Measure of Association in Single-Machine Mode F 79
The Simple Statistics table in Output 5.1.1 displays univariate descriptive statistics for the analysis variables.
By default, observations that have nonmissing values for each variable are used to derive the univariate
statistics for that variable.
Output 5.1.1 Simple Statistics
Measures of Association for a Physical Fitness Study

Execution Mode
Single-Machine
Number of Threads 4
3 Variables: Weight Oxygen RunTime
Simple Statistics
Variable
Mean Std Dev
Sum Minimum Maximum
Weight
31 77.44452 8.32857
2401 59.08000 91.63000
Oxygen
29 47.22721 5.47718
1370 37.38800 60.05500
RunTime 29 10.67414 1.39194 309.55000
8.17000 14.03000
The Pearson Correlation Coefficients table in Output 5.1.2 displays the Pearson correlation statistics for
pairs of analysis variables. The Pearson correlation is a parametric measure of association for two continuous
random variables. When the data have missing values, the number of observations used to calculate the
correlation can vary.
Output 5.1.2 Pearson Correlation Coefficients
Pearson Correlation Coefficients
Prob > |r| under H0: Rho=0
Number of Observations
Weight Oxygen RunTime
Weight
Oxygen
1.00000 -0.15358
0.4264
31
29
0.20072
0.2965
29
-0.15358 1.00000 -0.86843

0.4264
<.0001
29
29
28
RunTime 0.20072 -0.86843

0.2965 <.0001
29
28
1.00000
29
The table shows that the Pearson correlation between Runtime and Oxygen is 0.86843, which is significant
with a p-value less than 0.0001. This indicates a strong negative linear relationship between these two
variables. As Runtime increases, Oxygen decreases linearly.
Example 5.2: Computing the Pearson Measure of Association in Distributed

Mode
The real power of PROC HPCORR is when the computation is solved with multiple threads or in a distributed
environment.
You can switch to running in distributed mode simply by specifying valid values for the NODES=, INSTALL=,
and HOST= options in the PERFORMANCE statement.
An alternative to specifying the INSTALL= and HOST= options in the PERFORMANCE statement is to set
appropriate values for the GRIDHOST and GRIDINSTALLLOC environment variables by using OPTIONS
SET commands. For more information about setting these options or environment variables, see the section
Processing Modes on page 10 in Chapter 3, Shared Concepts and Topics.
The macro variable BRECLIB is the name of a libref to a billion-record database.
title 'PROC HPCORR Processing Billion-Record Database';
proc hpcorr data=&BRECLIB;
var x1-x5;
performance host=&GRIDHOST install=&GRIDINSTALLLOC;
run;
The execution mode in the Performance Information table shown in Output 5.2.1 indicates that the
calculations were performed in a distributed environment that uses 16 nodes; the data are predistributed using
a Greenplum parallel database.
Output 5.2.1 Performance Information in Distributed Mode
PROC HPCORR Processing Billion-Record Database

Host Node
Install Location
Execution Mode
Distributed
Another indication of distributed execution is the following message, which is issued by all high-performance
analytical procedures in the SAS log:
NOTE: The HPCORR procedure is executing in the distributed
computing environment with 16 worker nodes.
Because the sample database uses random data, the results are not meaningful. The power of highperformance analytics is that this test can be completed in a matter of minutes instead of hours.
References F 81
References
Anderson, T. W. (1984), An Introduction to Multivariate Statistical Analysis, 2nd Edition, New York: John
Wiley & Sons.
Blum, J. R., Kiefer, J., and Rosenblatt, M. (1961), Distribution Free Tests of Independence Based on the
Sample Distribution Function, Annals of Mathematical Statistics, 32, 485498.
Conover, W. J. (1999), Practical Nonparametric Statistics, 3rd Edition, New York: John Wiley & Sons.
Cox, N. R. (1974), Estimation of the Correlation between a Continuous and a Discrete Variable, Biometrics,
30, 171178.
Cronbach, L. J. (1951), Coefficient Alpha and the Internal Structure of Tests, Psychometrika, 16, 297334.
Drasgow, F. (1986), Polychoric and Polyserial Correlations, in S. Kotz, N. L. Johnson, and C. B. Read,
eds., Encyclopedia of Statistical Sciences, volume 7, 6874, New York: John Wiley & Sons.
Fisher, R. A. (1915), Frequency Distribution of the Values of the Correlation Coefficient in Samples from an
Indefinitely Large Population, Biometrika, 10, 507521.
Fisher, R. A. (1921), On the Probable Error of a Coefficient of Correlation Deduced from a Small Sample,
Metron, 1, 332.
Fisher, R. A. (1936), The Use of Multiple Measurements in Taxonomic Problems, Annals of Eugenics, 7,
179188.
Fisher, R. A. (1970), Statistical Methods for Research Workers, 14th Edition, Davien, CT: Hafner Publishing.
Hoeffding, W. (1948), A Non-parametric Test of Independence, Annals of Mathematical Statistics, 19,
546557.
Hollander, M. and Wolfe, D. A. (1999), Nonparametric Statistical Methods, 2nd Edition, New York: John
Wiley & Sons.
Keeping, E. S. (1962), Introduction to Statistical Inference, New York: D. Van Nostrand.
Knight, W. E. (1966), A Computer Method for Calculating Kendalls Tau with Ungrouped Data, Journal of
the American Statistical Association, 61, 436439.
Moore, D. S. (2000), Statistics: Concepts and Controversies, 5th Edition, New York: W. H. Freeman.
Mudholkar, G. S. (1983), Fishers z-Transformation, Encyclopedia of Statistical Sciences, 3, 130135.
Noether, G. E. (1967), Elements of Nonparametric Statistics, New York: John Wiley & Sons.
Novick, M. R. (1967), Coefficient Alpha and the Reliability of Composite Measurements, Psychometrika,
32, 113.
Nunnally, J. C. and Bernstein, I. H. (1994), Psychometric Theory, 3rd Edition, New York: McGraw-Hill.
Olsson, U., Drasgow, F., and Dorans, N. J. (1982), The Polyserial Correlation Coefficient, Biometrika, 47,
337347.
Ott, R. L. and Longnecker, M. T. (2000), An Introduction to Statistical Methods and Data Analysis, 5th
Edition, Belmont, CA: Wadsworth.
Spector, P. E. (1992), Summated Rating Scale Construction: An Introduction, Newbury Park, CA: Sage
Publications.
Yu, C. H. (2001), An Introduction to Computing and Interpreting Cronbach Coefficient Alpha in SAS,
in Proceedings of the Twenty-Sixth Annual SAS Users Group International Conference, Cary, NC: SAS
Institute Inc.
Chapter 6
The HPDMDB Procedure

Contents
Overview: HPDMDB Procedure . . . . . . . . . . . . . . . . . . .
Getting Started: HPDMDB Procedure . . . . . . . . . . . . . . . .
Syntax: HPDMDB Procedure . . . . . . . . . . . . . . . . . . . .
PROC HPDMDB Statement . . . . . . . . . . . . . . . . . .
CLASS Statement . . . . . . . . . . . . . . . . . . . . . . .
FREQ Statement . . . . . . . . . . . . . . . . . . . . . . . .
PERFORMANCE Statement . . . . . . . . . . . . . . . . .
VAR Statement . . . . . . . . . . . . . . . . . . . . . . . . .
WEIGHT Statement . . . . . . . . . . . . . . . . . . . . . .
Details: HPDMDB Procedure . . . . . . . . . . . . . . . . . . . .
Examples: HPDMDB Procedure . . . . . . . . . . . . . . . . . . .
Example 6.1: Running PROC HPDMDB on the Client . . . .
Example 6.2: Running with Client Data on the SAS appliance
Example 6.3: Running with Data on the SAS appliance . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
83
84
88
88
89
90
90
90
91
91
92
92
93
94
Overview: HPDMDB Procedure

The HPDMDB procedure is a high-performance version of the DMDB procedure, which creates summaries
of the input data source. PROC HPDMDB creates two output data sets: the VAROUT data set, which
contains a summary of the numeric variables, and the CLASSOUT data set, which contains a summary of the
classification variables.
PROC HPDMDB is high-performance in that it takes advantage of distributed and multicore computing
environments when the input data are stored on the SAS appliance.
You can use PROC HPDMDB to create a data mining database (DMDB) that is compatible with the DMDB
from PROC DMDB, although this feature of PROC HPDMDB might not be supported in future versions.
PROC HPDMDB runs in either single-machine mode or distributed mode.
Because the HPDMDB procedure is a high-performance analytical procedure, it also does the following:
computations
84 F Chapter 6: The HPDMDB Procedure
Topics.
Getting Started: HPDMDB Procedure

The HPDMDB procedure summarizes data. The following example uses the Sampsio.Hmeq data set, which
includes information about 5,960 fictitious mortgages. Each case represents an applicant for a home equity
loan, and all applicants have an existing mortgage. The binary target variable BAD indicates whether an
applicant eventually defaulted or was ever seriously delinquent. There are 10 numeric input variables and
three classification input variables.
proc hpdmdb data=Sampsio.Hmeq
classout=cout varout=vout;
var loan derog mortdue value yoj delinq
clage ninq clno debtinc;
class bad(desc) reason(ascending) job;
run;
proc print data=cout;run;
proc print data=vout;run;
The data set cout (shown in Figure 6.1) contains the class summary table with levels sorted according to
the sort option in the CLASS statement of PROC HPDMDB. You can see that the levels for BAD are in
descending order and the levels for REASON are in ascending order. The levels for JOB are in the default
ascending order.
Figure 6.1 Summaries of Classification Variables in Sampsio.Hmeq Data Set

Execution Mode
Single-Machine
Number of Threads 4
Data
Engine Role
WORK.COUT
V9
Output On Client
WORK.VOUT
V9
Output On Client
SAMPSIO.HMEQ V9
Input
Path
On Client
Getting Started: HPDMDB Procedure F 85

Obs NAME
LEVEL
CODE FREQUENCY TYPE CRAW
NRAW FREQPERCENT NMISSPERCENT
1 BAD
1189 N
19.9497
19.9497
2 BAD
4771 N
80.0503
80.0503
3 REASON
252 C
4.2282
4 REASON DEBTCON
3928 C
DebtCon
65.9060
68.8157
5 REASON HOMEIMP
1780 C
HomeImp
29.8658
31.1843
6 JOB
279 C
4.6812
7 JOB
MGR
767 C
Mgr
12.8691
13.5011
8 JOB
OFFICE
948 C
Office
15.9060
16.6872
9 JOB
OTHER
2388 C
Other
40.0671
42.0349
10 JOB
PROFEXE
1276 C
ProfExe
21.4094
22.4608
11 JOB
SALES
109 C
Sales
1.8289
1.9187
12 JOB
SELF
193 C
Self
3.2383
3.3973
Obs NAME
NMISS
1 LOAN
MIN
MAX
0 5960 1100.00
89900.00
MEAN
STD SKEWNESS KURTOSIS

2.02378
6.9326
0.85
5.32087
36.8728
73760.82 44457.61
1.81448
6.4819
112 5848 8000.00 855909.00 101776.05 57385.78
3.05334
24.3628
708 5252
3 MORTDUE
518 5442 2063.00 399550.00
4 VALUE
5 YOJ
515 5445
0.00
41.00
8.92
7.57
0.98846
0.3721
6 DELINQ
580 5380
0.00
15.00
0.45
1.13
4.02315
23.5654
7 CLAGE
308 5652
0.00
1168.23
179.77
85.81
1.34341
7.5995
8 NINQ
510 5450
0.00
17.00
1.19
1.73
2.62198
9.7865
222 5738
0.00
71.00
21.30
10.14
0.77505
1.1577
1267 4693
0.52
203.31
33.78
8.60
2.85235
50.5040
9 CLNO
10 DEBTINC
Obs
SUM
0.00
18607.97 11207.48
2 DEROG
USS
10.00
0.25
CSS
1 110903500.00 2.8121848E12 748495791434.56

2
1337.00
4099.00
3758.64
3 401406367.20 4.0362084E13 10754022449877

4 595186333.04 7.9830628E13 19254914800672
5
48581.75
745755.59
2418.00
7922.00
6835.25
1016038.99 224259958.52
41610414.32
312296.19
6464.00
23950.00
16283.34
122197.00
3192071.00
589751.93
10
158529.14
5702262.28
347161.26
Numeric summaries are in the data set vout, shown in Figure 6.2.
Figure 6.2 Summaries of Numeric Variables in Sampsio.Hmeq Data Set

Execution Mode
Single-Machine
Number of Threads 4
Data
Engine Role
WORK.COUT
V9
Output On Client
WORK.VOUT
V9
Output On Client
SAMPSIO.HMEQ V9
Obs NAME
LEVEL
Input
CODE FREQUENCY TYPE CRAW
Path
On Client
NRAW FREQPERCENT NMISSPERCENT
1 BAD
1189 N
19.9497
19.9497
2 BAD
80.0503
4771 N
80.0503
3 REASON
252 C
4.2282
4 REASON DEBTCON
3928 C
DebtCon
65.9060
68.8157
5 REASON HOMEIMP
1780 C
HomeImp
29.8658
31.1843
6 JOB
279 C
4.6812
7 JOB
MGR
767 C
Mgr
12.8691
13.5011
8 JOB
OFFICE
948 C
Office
15.9060
16.6872
9 JOB
OTHER
2388 C
Other
40.0671
42.0349
10 JOB
PROFEXE
1276 C
ProfExe
21.4094
22.4608
11 JOB
SALES
109 C
Sales
1.8289
1.9187
12 JOB
SELF
193 C
Self
3.2383
3.3973
Getting Started: HPDMDB Procedure F 87

Obs NAME
NMISS
1 LOAN
MIN
MAX
0 5960 1100.00
89900.00
MEAN
STD SKEWNESS KURTOSIS

2.02378
6.9326
0.85
5.32087
36.8728
73760.82 44457.61
1.81448
6.4819
112 5848 8000.00 855909.00 101776.05 57385.78
3.05334
24.3628
708 5252
3 MORTDUE
518 5442 2063.00 399550.00
4 VALUE
5 YOJ
515 5445
0.00
41.00
8.92
7.57
0.98846
0.3721
6 DELINQ
580 5380
0.00
15.00
0.45
1.13
4.02315
23.5654
7 CLAGE
308 5652
0.00
1168.23
179.77
85.81
1.34341
7.5995
8 NINQ
510 5450
0.00
17.00
1.19
1.73
2.62198
9.7865
222 5738
0.00
71.00
21.30
10.14
0.77505
1.1577
1267 4693
0.52
203.31
33.78
8.60
2.85235
50.5040
9 CLNO
10 DEBTINC
Obs
SUM
0.00
18607.97 11207.48
2 DEROG
USS
10.00
0.25
CSS
1 110903500.00 2.8121848E12 748495791434.56

2
1337.00
4099.00
3758.64
3 401406367.20 4.0362084E13 10754022449877

4 595186333.04 7.9830628E13 19254914800672
5
48581.75
745755.59
2418.00
7922.00
6835.25
1016038.99 224259958.52
41610414.32
312296.19
6464.00
23950.00
16283.34
122197.00
3192071.00
589751.93
10
158529.14
5702262.28
347161.26
Syntax: HPDMDB Procedure

The following statements are available in the HPDMDB procedure:
PROC HPDMDB DATA= < libref. >SAS-data-set < options > ;
CLASS variable (< order-option >) variable (< order-option >) . . . ;
FREQ variable ;
VAR variables ;
WEIGHT variable ;
PROC HPDMDB Statement

PROC HPDMDB DATA= < libref. >SAS-data-set < options > ;
The PROC HPDMDB statement invokes the procedure.
Required Arguments
DATA=< libref. >SAS-data-set
names the SAS data set that contains the information that you want added to the data mining database.
If the data set resides on the SAS appliance, then the SAS appliance is used during summarization.
Optional Arguments
DMDBCAT=< libref. >SAS-catalog
names the metadata catalog to be created by PROC HPDMDB.

CLASSOUT=< libref. >SAS-data-set
names the data set to contain the summaries of classification variables that are specified in the CLASS
statement.
VAROUT=< libref. >SAS-data-set
names the data set to contain the summaries of analysis variables that are specified in the VAR
statement.
VARDEF=divisor
specifies the divisor to use in the calculation of the variance and standard deviation. Table 6.1 shows
the possible values for divisor .
Table 6.1 Values for divisor
Value
Divisor
N
DF (default)
Number of observations (n)

Degrees of freedom (n 1)
CLASS Statement F 89
PRINT
prints the class-level information to all open ODS destinations.

SPECIALMISSING
enables special missing values to be treated as separate levels.

MAXLEVEL=max
specifies the maximum number of levels to be reported for each class variable. If more than max levels
of a class variable exist, PROC HPDMDB reports the frequency of unreported observations in a level
named _OTHER_.
CLASS Statement
CLASS variable (< order-option >) < variable (< order-option >) > . . . ;
The CLASS statement specifies the variables whose values define subgroup combinations for the analysis.
The CLASS and VAR statements are mutually exclusive.
Required Argument
variable
specifies a classification variable to be used in the analysis. For each level of a CLASS variable, the
CLASSOUT data set contains information about each of the following: the level value, its frequency,
and the type of the variable (numeric or character).
Optional Argument
order-option
specifies the order to use when considering the levels of CLASS variables to be sorted. The value of
order-option can be one of the following:
ASCENDING | ASC
arranges class levels in lowest-to-highest order of unformatted values.

DESCENDING | DESC
arranges class levels in highest-to-lowest order of unformatted values.

ASCFORMATTED | ASCFMT
arranges class levels in ascending order by their formatted values.

DESFORMATTED | DESFMT
arranges class levels in descending order by their formatted values.

DSORDER | DATA
arranges class levels according to the order of their appearance in the input data set.
N OTE : The DSORDER sort option is not supported for input data sets that are stored on the SAS
appliance.
FREQ Statement
FREQ variable ;
The FREQ statement specifies a numeric variable that contains the frequency of each observation.
Required Argument
variable
specifies a numeric variable whose value represents the frequency of the observation. For observations
where variable is 0 or missing, the observation is omitted in the CLASSOUT data set and is not
included in statistical calculations.
The PERFORMANCE statement defines performance parameters for multithreaded and distributed computing, passes variables about the distributed computing environment, and requests detailed results about the
performance characteristics of a SAS high-performance analytical procedure.
With the PERFORMANCE statement you can also control whether a SAS high-performance analytical
procedure executes in single-machine mode or distributed mode.
The PERFORMANCE statement for SAS high-performance analytical procedures is documented in the
section PERFORMANCE Statement on page 36 of Chapter 3, Shared Concepts and Topics.
VAR Statement
VAR variables ;
The VAR statement specifies one or more numeric variables to analyze. The CLASS and VAR statements are
mutually exclusive.
Required Argument
variables
specifies one or more numeric variables to be used in the analysis. The variables must be numeric. For
each variable, the VAROUT data set and the metadata contain the following statistics:
WEIGHT Statement F 91
Table 6.2 Statistics Recorded for VAR Variables
Statistic
Meaning
N
NMISS
MIN
MAX
MEAN
STD
SKEWNESS
KURTOSIS
SUM
CSS
USS
Number of observations that contain a missing value
Minimum observed value
Maximum observed value
Mean of observed values
Standard deviation
Measure of asymmetry
Measure of the heaviness of the tails
Sum of all nonmissing observations
Corrected sum of squares
Sum of squares
(See Appendix 1, SAS Elementary Statistics Procedures in Base SAS Procedures Guide, for formulas
and other details.)
WEIGHT Statement
WEIGHT variable ;
The WEIGHT statement specifies a numeric variable that contains a weight for each observation. The variable
is used in the computation of means, standard deviation, skewness, and kurtosis.
Required Argument
variable
represents how the observation should be weighted in statistical calculations. For observations where
variable is 0 or missing, the observation is still included in the CLASSOUT data set but the value is
not used in statistical calculations.
Details: HPDMDB Procedure

The statistics recorded for numeric variables are detailed in the section VAR Statement on page 90.
For classification variables, a level is a distinct observed value after formatting, removal of beginning and
ending white space, and capitalization. For example, the values MyLevel and MYLEVEL are treated as a single
level in the data set. Classification variables can be numeric, and the same levelization rules apply. For
example, 3.000002 and 3.0000001 are treated as the same level if they are formatted using BEST3.
Frequencies should be integers. If a noninteger frequency is specified, it is rounded to the nearest integer for
calculations. Weights do not need to be integers. Negative frequencies and weights are treated as 0.
Examples: HPDMDB Procedure

Example 6.1: Running PROC HPDMDB on the Client
This example demonstrates how to run the HPDMDB procedure on the following data set on the client:
data ex;
input x1 x2 x3 x4 y$ w f y2;
cards;
1 2 1 1 m .90 1 0
1 2 1 2 m .91 2 1
1 2 1 3 x .89 1 4
1 2 1 4 x .90 2 4
1 3 1 1 m .91 1 1
1 3 1 2 m .89 2 1
2 3 1 3 x .90 1 5
2 3 1 4 x .89 2 5
3 1 2 1 z .90 1 2
3 1 2 2 z .89 2 2
3 1 2 3 y .90 1 7
3 1 2 4 y .89 2 7
3 4 2 1 z .90 1 3
3 4 2 2 z .89 2 3
4 4 2 3 y .90 1 6
4 4 2 4 y .89 2 6
;
run;
When the input data set resides on the client and no PERFORMANCE statement is specified, as in the
following example, the client performs all computations:
proc hpdmdb data=ex print classout=cout varout=vout;
class x1-x3;
weight w;
var x4 y2;
freq f;
run;
Output 6.1.1 shows the summaries of the numeric variables in the data set ex.
Output 6.1.1 Summaries of Numeric Variables in ex Data Set
Obs NAME NMISS N MIN MAX
MEAN
STD SKEWNESS KURTOSIS SUM
1 x4
0 24
4 2.66326 1.06938
-0.05848
2 y2
0 24
7 3.57721 2.10658
0.12214
USS
CSS
-1.19599 57.26 178.80
26.302
-1.04243 76.91 377.19 102.067
Example 6.2: Running with Client Data on the SAS appliance F 93
Output 6.1.2 shows the summaries of the classification variables in the data set ex.
Output 6.1.2 Summaries of Classification Variables in ex Data Set
Obs NAME LEVEL CODE FREQUENCY TYPE CRAW NRAW FREQPERCENT NMISSPERCENT
1 x1
9 N
37.5
37.5
2 x1
3 N
12.5
12.5
3 x1
9 N
37.5
37.5
4 x1
3 N
12.5
12.5
5 x2
6 N
25.0
25.0
6 x2
6 N
25.0
25.0
7 x2
6 N
25.0
25.0
8 x2
6 N
25.0
25.0
9 x3
12 N
50.0
50.0
10 x3
12 N
50.0
50.0
Example 6.2: Running with Client Data on the SAS appliance

This example uses the same data set as in Example 6.1.
When the input data set resides on the client and a PERFORMANCE statement with a NODES= option is
specified, as in the following example, PROC HPDMDB copies the data set to the SAS appliance, which
does the computations. To run these statements successfully, you need to set the macro variables GRIDHOST
and GRIDINSTALLLOC to resolve to appropriate values, or you can replace the macro variable references in
the example with the appropriate values.
option set=GRIDHOST
= "&GRIDHOST";
option set=GRIDINSTALLLOC = "&GRIDINSTALLLOC";
/*Perform the computation on the SAS appliance using 5 nodes*/
proc hpdmdb data=ex print classout=cout varout=vout;
class x1-x3;
weight w;
var x4 y2;
freq f;
performance nodes=5 details;
run;
The results are the same as those shown in Output 6.1.1 and Output 6.1.2.
Example 6.3: Running with Data on the SAS appliance

When the input data set resides on the SAS appliance, the SAS appliance performs all computations and
reports the results back to the client. In the following example, the input data resides in the MyLib library,
which is a distributed data source. To run these statements successfully, you need to set the macro variables
to resolve to appropriate values, or you can replace the macro variable references with the appropriate values.
option set=GRIDHOST
= "&GRIDHOST";
libname MyLib &LIBTYPE
server ="&GRIDDATASERVER"
user
=&USER
password=&PASSWORD
database=&DATABASE;
/*MyLib is a libref for a distributed data source
In this case, the computation is automatically done
on the SAS appliance.*/
proc hpdmdb data=MyLib.ex print classout=cout varout=vout;
class x1-x3;
weight w;
var x4 y2;
freq f;
performance details;
run;
The results are the same as those shown in Output 6.1.1 and Output 6.1.2.
Chapter 7
The HPDS2 Procedure

Contents
Overview: HPDS2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
PROC HPDS2 Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
96
Single-Machine and Distributed Execution Modes . . . . . . . . . . . . . . . . . . .
97
Getting Started: HPDS2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
97
Syntax: HPDS2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
100
PROC HPDS2 Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
100
DATA Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
100
ENDDATA Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
101
101
QUIT Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
101
RUN Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
102
RUN CANCEL Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
102
Details: HPDS2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
102
Parallel Execution of DS2 Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
102
Limitations and Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
103
Packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
103
PERFORMANCE Statement Options . . . . . . . . . . . . . . . . . . . . .
103
Data Input/Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
103
Data Types and Declarations . . . . . . . . . . . . . . . . . . . . . . . . . .
103
Error Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
103
Examples: HPDS2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Example 7.1: Compute Mandelbrot Set . . . . . . . . . . . . . . . . . . . . . . . . .
104
104
Example 7.2: Aggregate Result Data Set . . . . . . . . . . . . . . . . . . . . . . . .
106
Overview: HPDS2 Procedure

The HPDS2 procedure enables you to submit DS2 language statements from a Base SAS session to one or
more machines in a grid for parallel execution. PROC HPDS2 verifies the syntactic correctness of the DS2
source on the client machine before submitting it for execution. The output data created by the DS2 DATA
statement can be output in either of the following ways: it can be written in parallel back to the grid data
store or it can be returned to the client machine and directed to any data store that is supported by SAS.
Because the DS2 code is executed in parallel on separate grid nodes that have single data partitions, each
node produces separate output that is the result of processing only the local data partition. As a result, it
96 F Chapter 7: The HPDS2 Procedure
might be necessary to use a second-stage program to aggregate the results from each node. The second stage
can be executed on the SAS client by using the DS2 procedure, where the SET statement reads all rows
created by the preceding parallel stage.
The syntax of DS2 is similar to that of the DATA step, but it does not include several key statements such as
INPUT and MERGE. In addition, using DS2 along with SAS high-performance analytical procedures limits
the PROC DS2 SET statement to a single input stream. The use of BY processing within the SET statement is
also not supported. Therefore, many of the traditional DATA step data preparation features are not available in
the HPDS2 procedure. PROC HPDS2 is most useful when significant amounts of computationally intensive,
row-independent logic must be applied to the data.
For more information about the DS2 language, see SAS DS2 Language Reference, which is available at
http://support.sas.com/documentation/solutions/ds2/DS2Ref.pdf.
PROC HPDS2 runs in either single-machine mode or distributed mode.
PROC HPDS2 Features

The HPDS2 procedure provides a vehicle for parallel execution of DS2 code in a distributed computing
environment. The following list summarizes the basic features of the HPDS2 procedure:
provides the ability to execute DS2 code in parallel
enables DS2 code to be executed on a local client machine (single-machine mode) or on multiple
machines in a distributed computing environment (distributed mode)
enables control of the level of parallelism per execution node and the number of nodes to engage
performs a syntax check of the DS2 code on the local client machine before sending it to the distributed
computing environment for execution
manages data migration to the location of execution and movement back to the client machine as
needed
Because the HPDS2 procedure is a high-performance analytical procedure, it also does the following:
computations
Topics.
Single-Machine and Distributed Execution Modes F 97
Single-Machine and Distributed Execution Modes

The HPDS2 procedure controls both the number of nodes that are engaged and the number of parallel
threads that each node uses for the execution of the DS2 language statements. In contrast to the THREADS
PACKAGE DS2 (whose syntax provides single-node scalability as part of the DS2 syntax), PROC HPDS2
provides threading that operates outside the syntax of the language.
In single-machine mode, one or more copies of the DS2 program can be executed in parallel in multiple
threads on the client machine.
In distributed mode, one or more copies of the DS2 program are executed in parallel on each machine in the
distributed computing environment. The distributed mode of execution has two variations:
In the client-data (local-data) model of distributed execution, the input data are not stored on the
appliance but are distibuted to the distributed computing environment by the SAS High-Performance
Analytics infrastructure during execution of the HPDS2 procedure.
In the alongside-the-database model of distributed execution, the data source is the database on the
appliance. The data are stored in the distributed database, and the DS2 program that is run on each node
is able to read and write the data in parallel during execution of the procedure. Instead of data being
moved across the network and possibly back to the client machine, data are passed locally between the
processes on each node of the appliance. In general, especially with large data sets, the best HPDS2
performance can be achieved if execution is alongside the database.
By default, the number of copies of the DS2 program that are executed in parallel on a given host (that is,
client machine or grid node) is determined by the HPDS2 procedure based on the number of CPUs (cores)
available on the host machine. The default is to execute one instance of the DS2 program in a dedicated
thread per CPU. You can change the default by specifying the NTHREADS= option in the PERFORMANCE
statement. For example, if you specify NTHREADS=n, then the HPDS2 procedure runs n copies of the DS2
program in parallel on each machine.
For information about the available modes of execution and how to switch between them, see the section
Processing Modes on page 10 in Chapter 3, Shared Concepts and Topics.
Getting Started: HPDS2 Procedure

This example illustrates a simple HPDS2 procedure. In this case, the DS2 source statements are executed
alongside the database in distributed mode. The DS2 code that is submitted to the grid is contained within
the DATA and ENDDATA statements. The following DATA step creates a data set that consists of fictitious
daily temperatures that are collected from a number of U.S. airports during a period of one week:
data daily_temps;
input city $ mon tue wed thu fri;
datalines;
lax 88 92 94 97 86
sfo 65 60 75 72 74
nyc 99 95 94 95 90
phl 92 89 91 93
atl 95 99 92 98
den 85 87 89 72
pit 68 70 72 73
rdu 98 95 99 95
dtt 88 90 90 87
anc 51 56 60 64
sea 72 78 77 80
msy 98 97 99 98
mia 95 92 98 94
ord 87 85 84 80
dfw 95 96 97 95
hou 98 99 98 97
las 104 105 102
pdx 78 82 76 74
san 80 81 78 82
phx 95 98 95 97
cle 75 72 76 80
ont 78 80 78 81
tpa 94 94 92 90
bos 80 78 77 75
clt 83 80 79 80
;
run;
94
94
73
77
96
92
62
79
99
96
79
97
92
99 101
80
80
100
78
72
92
79
81
The HPDS2 procedure reads this data set and calculates a daily average temperature in Fahrenheit and Celsius
for each airport and then provides a synopsis of the weekly temperature average.
In the following statements, the driver DS2GTF.out in the DATA statement refers to the input data set, and
the SET DS2GTF.in statement refers to the output data set:
libname applianc
server =
user
=
password
database
&ENGINE
"&GRIDDATASERVER"
&USER
= &PASSWORD
= &DATABASE;
proc hpds2 data=daily_temps

out=applianc.avg_temps;
performance host="&GRIDHOST" install="&GRIDINSTALLLOC";
data DS2GTF.out;
dcl double avgf avgc;
dcl char(5) synopsis;
method run();
set DS2GTF.in;
avgf = mean(mon, tue, wed, thu, fri);
avgc = round((avgf - 32.0) * 5.0/9.0, .1);
if avgf >= 95.0 then synopsis = 'Hot';
else if avgf > 80.0 then synopsis = 'Warm';
else if avgf > 60.0 then synopsis = 'Mild';
else synopsis = 'Cold';
end;
enddata;
run;
Getting Started: HPDS2 Procedure F 99
The following PRINT procedure displays the table of average temperatures that are produced by the HPDS2
procedure:
proc print data=applianc.avg_temps;
title1 'Average Temperatures';
var city synopsis avgf avgc;
run;
Figure 7.1 displays the output of the PRINT procedure.

Figure 7.1 Average Temperatures
Average Temperatures
Obs city synopsis avgf avgc
1 lax
Warm
91.4 33.0
2 sfo
Mild
69.2 20.7
3 nyc Warm
94.6 34.8
4 phl
Warm
91.8 33.2
5 atl
Hot
95.6 35.3
6 den Warm
81.2 27.3
7 pit
Mild
72.0 22.2
8 rdu Hot
96.6 35.9
9 dtt
89.4 31.9
Warm
10 anc Cold
58.6 14.8
11 sea Mild
77.2 25.1
12 msy Hot
98.2 36.8
13 mia Hot
95.0 35.0
14 ord Warm
83.0 28.3
15 dfw Hot
96.0 35.6
16 hou Hot
96.8 36.0
17 las
Hot
102.2 39.0
18 pdx Mild
78.0 25.6
19 san Warm
80.2 26.8
20 phx Hot
97.0 36.1
21 cle
Mild
76.2 24.6
22 ont Mild
77.8 25.4
23 tpa Warm
92.4 33.6
24 bos Mild
77.8 25.4
25 clt
80.6 27.0
Warm
Syntax: HPDS2 Procedure

The following statements are available in the HPDS2 procedure:
PROC HPDS2 < options > ;
DATA DS2GTF.out ;
DS2 statements
METHOD RUN()
SET DS2GTF.in
END
ENDDATA ;
RUN ;
RUN CANCEL ;
QUIT ;
PROC HPDS2 Statement

PROC HPDS2 < options > ;
The PROC HPDS2 statement invokes the procedure.

You can specify the following options in the PROC HPDS2 statement:
DATA=SAS-data-set
IN=data-set
names the SAS data set or database table to be used by PROC HPDS2. The default is the most recently
created data set.
OUTPUT=data-set
OUT=data-set
names the SAS data set or database table that is created by PROC HPDS2.
FMTLIBXML=file-ref
specifies the file reference for a format stream.
DATA Statement
DATA DS2GTF.out ;
The DATA statement indicates the beginning of the DS2 code block. The code block terminates with the
ENDDATA statement.
ENDDATA Statement F 101
A reference to the DS2 driver (DS2GTF.out) must be included as part of the DATA statement. If an input
data set is specified in the PROC HPDS2 statement, then a run() method must be included in the DS2 code
block and the first statement after the METHOD RUN() statement must be the SET DS2GTF.in statement.
DS2GTF.out and DS2GTF.in refer to the input and output data sets, respectively.
ENDDATA Statement
ENDDATA ;
The ENDDATA statement terminates the DS2 code block. The statements between the DATA and ENDDATA
statement are submitted to the grid for execution. The DS2 run, init, and term methods are specified between
the DATA and ENDDATA statements.
the performance characteristics of PROC HPDS2.
You can also use the PERFORMANCE statement to control whether PROC HPDS2 executes in singlemachine or distributed mode.
It is important to remember the distinction between the NODES= and NTHREADS= options. The NODES=
option specifies the number of separate distributed nodes that participate in the DS2 execution, whereas the
NTHREADS= option determines how many independent copies of the DS2 program are run in parallel on
each node. If the data are located on the grid, then all nodes must be engaged; therefore, the NODES= option
might be overridden. Setting NODES=0 causes the DS2 code to execute on the client side only. Setting the
NTHREADS= option to a value that is greater than the CPU count on each grid node is not likely to improve
overall throughput.
For more information about the PERFORMANCE statement, see the section PERFORMANCE Statement
on page 36 of Chapter 3, Shared Concepts and Topics.
QUIT Statement
QUIT ;
The QUIT statement stops the procedure. PROC HPDS2 statements that have not been submitted by a RUN
statement are terminated.
RUN Statement
RUN ;
The RUN statement submits the PROC HPDS2 statements that precede it for execution.
PROC HPDS2 requires the RUN statement to submit the statements. That is, SAS reads the program
statements that are associated with one task until it reaches a RUN statement.
RUN CANCEL Statement

RUN CANCEL ;
The RUN CANCEL statement cancels the PROC HPDS2 statements that precede it. RUN CANCEL is useful
if you enter a typographical error.
Details: HPDS2 Procedure

Parallel Execution of DS2 Code
An important characteristic of multithreaded or distributed applications is that they might produce nondeterministic or unpredictable results. The exact behavior of a DS2 program running in parallel on the grid is
influenced by a number of factors, including the pattern of data distribution that is used, the execution mode
that is chosen, the number of compute nodes and threads that are used, and so on. The HPDS2 procedure
does not examine whether the DS2 code that is submitted produces meaningful and reproducible results.
It simply executes the DS2 code that is provided on each of the units of work, whether these are multiple
threads on a single machine or multiple threads on separate grid nodes. Each instance of the DS2 program
operates on a subset of the data. The results that are produced by each unit of work are then gathered, without
further aggregation, into the output data set.
Because the DS2 code instances are executed in parallel, consideration must be given to the DS2 language
elements that are included in the DS2 code block of an HPDS2 procedure. Not all DS2 language elements
can be meaningfully used in multithreaded or distributed applications. For example, lagging or retaining
of variables can imply ordering of observations. A deterministic order of observations does not exist in
distributed applications, and enforcing data order might have a negative impact on performance.
Optimal performance is achieved when the input data are stored in the distributed database and the grid host
is the appliance that houses the data. With the data distributed in this manner, the different instances of the
DS2 code running on the grid nodes can read the input data and write the output data in parallel from the
local database management system (DBMS).
Limitations and Issues F 103
Limitations and Issues

The current release of the HPDS2 procedure does not support all of the features of the DS2 language. The
following subsections summarizes the known limitations and issues for PROC HPDS2.
Packages
DS2 packages are collections of variables and methods that can be used in DS2 programs and threads. The
HPDS2 procedure does not support DS2 packages at this time. Use of the PACKAGE and ENDPACKAGE
constructs within an HPDS2 procedure results in an error.
PERFORMANCE Statement Options

Setting the NTHREADS= option in the PERFORMANCE statement to very high values can cause out-ofmemory errors. For example, out-of-memory errors have been seen with NTHREADS=100.
Data Input/Output
If an input data set is specified, then a SET DS2GTF.in statement must be included in the METHOD RUN()
statement. If either the SET DS2GTF.in or the SET DS2GTF.out driver reference is missing, then the SAS
session stops responding.
The use of BY groups within the SET statement of an HPDS2 procedure is not supported at this time.
When used within an HPDS2 procedure, the PUT statement does not currently write any data to the client
log.
The OVERWRITE= option is not supported in PROC HPDS2.
Data Types and Declarations

The HPDS2 procedure does not support the following data types: REAL, TINYINT, NCHAR, TIMESTAMP,
DATE, and TIME. If any of these data types are declared within an HPDS2 procedure, then an error is
displayed.
Informats are not currently supported in PROC HPDS2.
Delimited identifiers (for example, dcl double "a%& b") are not currently supported in PROC HPDS2.
No warning or error messages are output when assignments that involve out-of-bounds arrays are used within
an HPDS2 procedure.
Error Messages
Incorrect source line numbers are reported when there is an error in an HPDS2 procedure. In addition, the
ordering of error messages displayed is reversed for PROC HPDS2 from the order of error messages that is
output for DS2.
Examples: HPDS2 Procedure

Example 7.1: Compute Mandelbrot Set
This example computes and plots a Mandelbrot set. The DS2 source statements that compute the set of
coordinates that comprise the Mandelbrot set are submitted to the grid and executed alongside the database
in distributed mode. Note that Mandelbrot set computation is perfectly scalable in that each point can be
computed independently of every other point.
This example uses a DS2 procedure to create a data set that consists of one row for each Mandelbrot
coordinate to be computed. The HPDS2 procedure reads this data set and computes the coordinates. The
Mandelbrot set is then graphed by using the GCONTOUR procedure.
libname applianc
server =
user
=
password
database
&ENGINE
"&GRIDDATASERVER"
&USER
= &PASSWORD
= &DATABASE;
/* Set up the table that contains one row for each coordinate to compute */
proc ds2;
data inp(overwrite=yes);
dcl double p q r;
dcl integer maxiterate;
method init();
dcl int n m;
dcl int i j k;
dcl double pmin pmax qmin qmax;
n = 1024;
m = 1024;
pmin = -1.5; pmax = -0.5;
qmin = -0.5; qmax = 0.5;
r = 100.0;
maxiterate = 50;
do k = 1 to n*m;
i = k/m;
j = mod(k,m);
p = i*(pmax-pmin)/(n-1)+pmin;
q = j*(qmax-qmin)/(m-1)+qmin;
output;
end;
end;
enddata;
run;
quit;
/* Compute the coordinates */
proc hpds2 data=inp out=applianc.mandelbrot;
data DS2GTF.out;
Example 7.1: Compute Mandelbrot Set F 105
dcl int mesh;

dcl double x y rr nx ny;
keep p q mesh;
method run();
set DS2GTF.in;
x = p;
y = q;
rr = r**2;
mesh = 0;
do while (mesh < maxiterate and (x**2+y**2 < rr));
nx = x**2 - y**2 + p;
ny = 2.0*x*y + q;
x = nx;
y = ny;
mesh = mesh+1;
end;
end;
enddata;
run;
/* Plot the results */
goptions colors= (
CX003366 CX336699 CX6699CC
CX336600 CX669933 CX99CC66
CX660033 CX993366 CXCC6699
CX003366 CX663399 CX9966CC
CX003366 CX663399 CX9966CC
black
)
;
CX99CCFF
CXCCFF99
CXFF99CC
CXCC99FF
CXCC99FF
CX006633
CX663300
CX003366
CX003366
CX003366
CX339966
CX996633
CX663399
CX663399
CX663399
CX66CC99
CXCC9966
CX9966CC
CX9966CC
CX9966CC
CX99FFCC
CXFFCC99
CXCC99FF
CXCC99FF
CXCC99FF
proc gcontour data=applianc.mandelbrot;

Title 'Mandelbrot Set';
plot q*p=mesh /
nolegend
pattern
join
levels = 5 to 45
;
run;
Output 7.1.1 shows the graphic representation of the Mandelbrot set that is computed by the HPDS2
procedure.
Output 7.1.1 Computed Mandelbrot Set
Example 7.2: Aggregate Result Data Set

This example illustrates how the intermediate result data that are generated from the DS2 code running in
parallel on separate grid nodes can be aggregated into a final result data set. In this case, the aggregation is
done by a second-stage PROC DS2 program that executes on the SAS client.
This example uses a DATA step program that runs on the SAS client to generate a sample data set that
consists of dimensional information for each of 200 objects (closed cylinders). These data are used by the
HPDS2 procedure to calculate the volume and surface area of each object. The second-stage DS2 procedure
aggregates these results, summing the total volume and surface area for all objects and computing the average
volume and surface area. In this example, the DS2 code running in parallel on the grid is used to perform
the row-independent and computationally intensive portion of the processing, whereas the work done by the
second-stage DS2 procedure is limited to the final result aggregation and summary.
libname applianc
server =
user
=
password
database
&ENGINE
"&GRIDDATASERVER"
&USER
= &PASSWORD
= &DATABASE;
Example 7.2: Aggregate Result Data Set F 107
data obj_dims;
do id=1 to 200;
radius = ranuni(1) * 10;
height = ranuni(1) * 20;
output;
end;
run;
%let pi=3.14159;
proc hpds2 data=obj_dims
out=applianc.obj_comps;
data DS2GTF.out;
method run();
set DS2GTF.in;
volume = &pi * radius**2 * height;
area = (2 * &pi * radius**2) + (2 * &pi * radius * height);
end;
enddata;
run;
proc print data=applianc.obj_comps (obs=20);
title1 'Volumes and Areas';
run;
data obj_comps;
set applianc.obj_comps;
run;
proc ds2;
data obj_totals (keep = (ncount vsum asum vmean amean));
dcl double ncount vsum asum vmean amean;
method init();
ncount = 0;
vsum = 0;
asum = 0;
end;
method run();
set {select volume, area from obj_comps};
ncount + 1;
vsum + volume;
asum + area;
end;
method term();
if ncount ne 0 then do;
vmean = vsum/ncount;
amean = asum/ncount;
end;
output;
end;
enddata;
run;
quit;
proc print data=obj_totals;

title1 'Total Volume and Area';
run;
Output 7.2.1 shows a subset of the volumes and areas that are computed by the HPDS2 procedure.
Output 7.2.1 Computed Volumes and Areas
Volumes and Areas

Obs id
radius
height volume
area
1 1.8496256982 19.401774313
208.53
246.97
2 3.9982430609 5.1879729075
260.55
230.77
3 9.2160257787 19.385546995 5172.67 1656.20
4 5.4297917315 10.633834456
5 0.4979402621
1.04
5.72
6 8.1931857058 10.477410429 2209.58
961.15
7 8.5339431085 1.3436915359
307.43
529.64
8 9.5702385761
5.943879283 1710.27
932.89
9 2.7261178907 13.798592619
1.331331032
984.93
548.03
322.16
283.05
10 10 9.7676486241 4.5301503709 1357.82
877.48
11 11 6.8823655028 8.2552773264 1228.45
654.60
12 12 5.5855411271 5.7445122142
563.03
397.63
13 13 4.7578930504
16.89973954 1201.87
647.45
14 14 6.3452411845 11.807293385 1493.47
723.71
15 15 5.8258152641 7.5402673835
803.99
489.26
16 16 7.2836155991 10.132070589 1688.66
797.02
17 17 9.3121359401 18.582400996 5062.32 1632.10

18 18 5.8966033794 5.9444569265
649.33
438.70
19 19
3.910424334 9.4485835123
453.90
328.23
20 20 6.7952574821 3.3617670198
487.67
433.66
Output 7.2.2 shows the aggregated results that are produced by the second-stage DS2 program.
Output 7.2.2 Computed Total Volume and Area
Total Volume and Area

Obs ncount
1
vsum
asum vmean amean
200 209883.99 104680.26 1049.42 523.401
Chapter 8
The HPIMPUTE Procedure

Contents
Overview: HPIMPUTE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
109
PROC HPIMPUTE Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
110
110
Getting Started: HPIMPUTE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . .
111
Syntax: HPIMPUTE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
112
PROC HPIMPUTE Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
112
CODE Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
113
FREQ Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
113
ID Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
113
IMPUTE Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
113
INPUT Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
114
114
Details: HPIMPUTE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
115
Obtaining the Statistics for Imputation . . . . . . . . . . . . . . . . . . . . . . . . .
115
Random Imputation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
115
Displayed Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
116
ODS Table Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
116
Examples: HPIMPUTE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
117
Example 8.1: Running Client Data on the Client . . . . . . . . . . . . . . . . . . . .
117
Example 8.2: Running Client Data on the SAS Appliance . . . . . . . . . . . . . . .
117
Overview: HPIMPUTE Procedure

The HPIMPUTE procedure executes high-performance numeric variable imputation. Imputation is a common
step in the data preparation stage. The HPIMPUTE procedure takes only numeric variables.
The HPIMPUTE procedure can replace numeric missing values with a specified value. It can also replace
numeric missing values with the mean, the pseudomedian, or some random value between the minimum
value and the maximum value of the nonmissing values.
When PROC HPIMPUTE calculates the mean, the pseudomedian, or a random value, it ignores any observation that has a FREQ variable whose value is less than or equal to 0.
The HPIMPUTE procedure runs in either single-machine mode or distributed mode.
110 F Chapter 8: The HPIMPUTE Procedure
PROC HPIMPUTE Features

PROC HPIMPUTE provides a vehicle for the parallel execution of imputation. The following list summarizes
the basic features of PROC HPIMPUTE:
provides the ability to execute imputation
needed
Because the HPIMPUTE procedure is a high-performance analytical procedure, it also does the following:
computations
Topics.

The HPIMPUTE procedure can exploit computer grids by imputing independently on different grid nodes
in parallel, and it supports multithreading on each node. For more information about single-machine and
distributed execution modes, see the section Processing Modes on page 10 in Chapter 3, Shared Concepts
and Topics.
You can control both the number of parallel threads per execution node and the number of computing nodes
to engage.
Alternatively, PROC HPIMPUTE can be executed on a grid of distributed computers. In distributed mode,
one or more copies of the imputation code are executed in parallel on each grid node.
The distributed mode of execution has two variations:
In the client-data (local-data) model of distributed execution, the input data are not stored on the appliance but are distributed to the distributed computing environment during execution of the HPIMPUTE
procedure.
appliance. The data are stored in the distributed database, and the imputation code that runs on each
node can read and write the data in parallel during execution of the HPIMPUTE procedure. Instead
of being moved across the network and possibly back to the client machine, data are passed locally
between the processes on each node of the appliance. In general and especially with large data sets, the
best PROC HPIMPUTE performance can be achieved if execution is alongside the database.
Getting Started: HPIMPUTE Procedure F 111
Getting Started: HPIMPUTE Procedure

The HPIMPUTE procedure can use four methods to impute numeric missing values. This example uses
all four imputation methods available in the IMPUTE statement to manipulate a data set. The following
SAS DATA step creates the SAS data set ex1, which has six variables: the first four variables all have some
missing values, the fifth variable is the frequency variable, and the last variable is an index variable.
data ex1;
input a b
cards;
2
2
.
2
2
.
.
2
.
2
run;
c
3
2
0
3
.
6
4
5
6
3
freq id;
1
2
3
.
.
.
.
.
9
10
1
2
.
.
.
.
.
.
9
10
2
3
0
.
-5
3
4
3
1
3
1
2
3
4
5
6
7
8
9
10
The following statements include four IMPUTE statements, each of which specifies a different imputation
method. The INPUT statement specifies the input variables. PROC HPIMPUTE assumes that the variables
have an interval level of measurement because the variables are numeric.
proc hpimpute data=ex1 out=out1;
id id;
input a b c d;
impute a / value=0.1;
impute b / method=pmedian;
impute c / method=random;
impute d / method=mean;
freq freq;
run;
Output 8.1 shows the imputation results. The Variable column shows the original variable names from the
input data set. The Imputation Indicator column shows a 0 if that observation was not imputed and 1 if
it was. The Imputed Variable column shows the names of imputed variables in the output data set. The
Type of Imputation column shows the types of imputation methods: Given Value, Pseudo Median, Random
(between the minimum value and the maximum value of the nonmissing values), and Mean. For random
imputation, the last column shows the imputation seed. For other imputation methods, the last column shows
the imputation value that is used to replace missing values.
Figure 8.1 HPIMPUTE Getting Started Example Output

Imputation Results
Imputation Imputed
N Type of
Variable Indicator Variable Missing Imputation
Imputation
Value (Seed)
M_a
IM_a
4 Given value
0.10000
M_b
IM_b
1 Pseudo Median
4.00000
M_c
IM_c
5 Random
5.00000
M_d
IM_d
6 Mean
5.22222
Syntax: HPIMPUTE Procedure

The following statements are available in the HPIMPUTE procedure:
PROC HPIMPUTE < options > ;
INPUT variables ;
IMPUTE variables < options > ;
ID variables ;
FREQ variable ;
CODE < options > ;
The PROC HPIMPUTE statement, INPUT statement, and IMPUTE statement are required. The INPUT
statement and IMPUTE statement can appear multiple times.
PROC HPIMPUTE Statement

PROC HPIMPUTE < options > ;
The PROC HPIMPUTE statement invokes the procedure. You can specify one or both of the following
options:
DATA=< libref. >SAS-data-set
names the SAS data set for which PROC HPIMPUTE is to impute values. The default is the most
recently created data set. If the data are already distributed, PROC HPIMPUTE reads the data alongside
the distributed database. For information about the various execution modes and about the alongsidethe-database model, see the sections Processing Modes on page 10 and Alongside-the-Database
Execution on page 18 in Chapter 3, Shared Concepts and Topics.. WHERE processing is supported.
OUT=< libref. >SAS-data-set
names the SAS data set to which PROC HPIMPUTE writes the data along with ID variables (if
applicable), imputation indicator variables (0 for not imputed or 1 for imputed), and imputed variables.
There is no default output data set.
CODE Statement F 113
CODE Statement
CODE < options > ;
The CODE statement generates SAS DATA step code that mimics the computations that are performed when
the IMPUTE statement runs in single-machine mode and uses a single thread. You can specify the following
options:
FILE=filename
specifies the name of the file to which the SAS score code is written.
FREQ Statement
FREQ variable ;
The variable in the FREQ statement identifies a numeric variable in the data set that contains the frequency
of occurrence for each observation. PROC HPIMPUTE treats each observation as if it appeared n times,
where n is the value of the FREQ variable for the observation. If the frequency value is not an integer, it is
truncated to an integer. If the frequency value is less than 1 or missing, the observation is not used in the
analysis. When the FREQ statement is not specified, each observation is assigned a frequency of 1.
ID Statement
ID variables ;
The optional ID statement lists one or more variables from the input data set that are transferred to the output
data set. The ID statement accepts numeric and character variables. For example, when you specify the
OUT= option in the PROC HPIMPUTE statement, the ID variables, followed by indicator variables and
imputed variables, are added to the output data set.
By default, to avoid data duplication for large data sets, the HPIMPUTE procedure does not include any
variables from the input data set in the output data sets. Therefore, the ID statement can be used to copy
variables from the input data set to the output data set.
The variables in an ID statement must not appear in any INPUT statement; if they appear, an error is reported.
IMPUTE Statement
IMPUTE variables < options > ;
The IMPUTE statement names the variables whose values PROC HPIMPUTE imputes. You can specify
multiple IMPUTE statements. The IMPUTE statement takes only numeric variables; character variables are
reported as an error. The variables in an IMPUTE statement must appear in an INPUT statement; if they do
not appear in an INPUT statement, an error is reported.
You can specify the following options:

METHOD=MEAN | RANDOM | PMEDIAN
specifies the method of imputation. You can specify the following values:
MEAN
replaces missing values with the algebraic mean of the variable. If there is no
nonmissing value for the variable to be imputed, the imputation result remains
missing.
RANDOM
replaces missing values with a random value that is drawn between the minimum
and maximum of the variable. If there is no nonmissing value for the variable to be
imputed, the imputation result remains missing.
PMEDIAN
replaces missing values with the pseudomedian of the variable. If there is no

nonmissing value for the variable to be imputed, the imputation result remains
missing.
N OTE : If you specify the method of imputation and all the values for all the variables to be imputed
are missing, PROC HPIMPUTE exits with an error.
VALUE=value
replaces missing values with the specified value.
INPUT Statement
INPUT variables ;
The INPUT statement names one or more input variables. You can specify multiple INPUT statements. The
INPUT statement takes only numeric variables; character variables are reported as an error.
the performance characteristics of the HPIMPUTE procedure.
You can also use the PERFORMANCE statement to control whether the HPIMPUTE procedure executes in
Details: HPIMPUTE Procedure F 115
Details: HPIMPUTE Procedure

Obtaining the Statistics for Imputation
PROC HPIMPUTE first computes the imputation value and then imputes with that value. Some statistics
(such as the mean, minimum, and maximum) are computed precisely. The pseudomedian, which is calculated
if you specify METHOD=PMEDIAN in the IMPUTE statement, is an estimation of the median. The
computation of the median requires sorting the entire data. However, in a distributed computing environment,
each grid node contains only a part of the entire data. Sorting all the data in such an environment requires
a lot of internode communications, degrading the performance dramatically. To address this challenge, a
binning-based method is used to estimate the pseudomedian.
For variable x , assume that the data set is {xi }, where i D 1; 2; :::; n. Let mi n.x/ D mini 2f1::ng fxi g, and
let max.x/ D maxi 2f1::ng fxi g. The range of the variable is range.x/ D max.x/ mi n.x/.
A simple bucket binning method is used to obtain the basic information. Let N be the number of buckets,
ranging from mi n.x/ to max.x/. For each bucket Bi , i D 1; 2; :::; N , PROC HPIMPUTE keeps following
information:
ci : count of x in Bi
mi ni : minimum value of x in Bi
maxi : maximum value of x in Bi
For each bucket Bi , the range Ri is
(
mi n.x/ C .i 1/ d; mi n.x/ C i d /
Ri D
mi n.x/ C .i 1/ d; max.x/
where d D
if i < N
if i D N
max.x/ mi n.x/
N
P
To calculate the pseudomedian, PROC HPIMPUTE finds the smallest I , such that IiD1 ci >D 0:5 m,
where m is the number of nonmissing observations of x in the data set. Therefore, the pseudomedian value Q
is
(
P
mi nI if IiD1 ci > 0:5 m
QD
P
maxI if IiD1 ci D 0:5 m
N is set to 10,000 in PROC HPIMPUTE. Experiments show that the pseudomedian is a good estimate of the
median and that the performance is satisfactory.
Random Imputation
If you specify METHOD=RANDOM in the IMPUTE statement, PROC HPIMPUTE replaces missing
values with a random value that is drawn between the minimum and maximum of the variable. For
variable x , assume that the data set is {xi }, where i D 1; 2; :::; n. Let mi n.x/ D mini 2f1::ng fxi g, and let
max.x/ D maxi 2f1::ng fxi g. The random value R D mi n.x/ C .max.x/ mi n.x// ranuni.SEED/,
where ranuni./ is a function that takes a SEED (number) as input and returns a random value from a
uniform distribution between 0 and 1. When PROC IMPUTE runs in single-machine mode and uses a single
thread, the SEED is set to 5. When PROC IMPUTE runs in distributed execution mode or uses multiple
threads, the SEED is determined at run time.
Displayed Output
The HPIMPUTE procedure displays imputation results and performance information.
The Imputation Results table includes six columns. The first column shows the original variable names
from the input data set. The second column displays a 0 if that observation was not imputed and a 1 if it was.
The third column shows the names of imputed variables in the output data set. The fourth column shows the
number of missing values. The fifth column shows the types of imputation methods: Given Value, Mean,
Pseudo Median, Random (between the minimum value and the maximum value of the nonmissing values).
The last column shows the imputation seed for which random imputation generated the imputation value, or
the imputation value for other imputation methods that replace missing values.
The Performance Information table is produced by default. It displays information about the execution
mode. For single-machine mode, the table displays the number of threads used. For distributed mode, the
table displays the grid mode (symmetric or asymmetric), the number of compute nodes, and the number of
threads per node.
If you specify the DETAILS option in the PERFORMANCE statement, the procedure also produces a
Timing table in which elapsed times (absolute and relative) for the main tasks of the procedure are
displayed.
ODS Table Names

Table 8.1 lists the names of the data tables that are created by the HPIMPUTE procedure. You must use these
names in ODS statements.
Table 8.1 ODS Tables Produced by PROC HPIMPUTE
Table Name
Description
Required Statement or Option
PerformanceInfo
Performance information
Default output
ImputeResults
Imputation results
Default output
Timing
Timing
PERFORMANCE statement with

DETAILS option
Example 8.2: Running Client Data on the SAS Appliance F 117
Examples: HPIMPUTE Procedure

Example 8.1: Running Client Data on the Client
This example demonstrates how to use PROC HPIMPUTE to perform imputation on the Sampsio.Hmeq data
set, which resides on the client.
following statements, the client performs all computations:
/*sampsio is a libref for a data source on the client.*/
proc hpimpute data=sampsio.hmeq out=out1;
input mortdue value clage debtinc;
impute mortdue / value = 70000;
impute value
/ method = mean;
impute clage
/ method = random;
impute debtinc / method = pmedian;
run;
Output 8.1.1 shows the imputation results.

Output 8.1.1 Imputation Results

Imputation Results
Variable
Imputation
Indicator
Imputed
Variable
N Type of
Missing Imputation
Imputation
Value (Seed)
MORTDUE M_MORTDUE IM_MORTDUE
518 Given value
VALUE
M_VALUE
IM_VALUE
112 Mean
101776
CLAGE
M_CLAGE
IM_CLAGE
308 Random
5.00000
DEBTINC
M_DEBTINC IM_DEBTINC
1267 Pseudo Median
70000
34.81696
Output 8.1.2 shows the performance information.

Output 8.1.2 Performance Information

Execution Mode
Single-Machine
Number of Threads 4
Example 8.2: Running Client Data on the SAS Appliance

When the input data set resides on the client and a PERFORMANCE statement that includes a NODES=
option is specified, as in the following statements, PROC HPIMPUTE copies the data set to the SAS appliance,
where the imputation is performed:

option set=GRIDHOST="&GRIDHOST";
option set=GRIDINSTALLLOC="&GRIDINSTALLLOC";
proc hpimpute data=sampsio.hmeq out=out2;
input mortdue value clage debtinc;
impute mortdue / value = 70000;
impute value
/ method = mean;
impute clage
/ method = random;
impute debtinc / method = pmedian;
performance nodes=2 details
run;
Output 8.2.1 shows the imputation results.

Output 8.2.1 Imputation Results

Imputation Results
Variable
Imputation
Indicator
Imputed
Variable
N Type of
Missing Imputation
Imputation
Value (Seed)
MORTDUE M_MORTDUE IM_MORTDUE
518 Given value
VALUE
M_VALUE
IM_VALUE
112 Mean
101776
CLAGE
M_CLAGE
IM_CLAGE
308 Random
5.00000
DEBTINC
M_DEBTINC IM_DEBTINC
1267 Pseudo Median
Output 8.2.2 shows the performance information.

Host Node
Install Location
Execution Mode
Distributed
Output 8.2.3 shows the timing information.

Output 8.2.3 Procedure Task Timing
Procedure Task Timing
Task
Seconds Percent
Startup of Distributed Environment
2.80 96.79%
Data Transfer from Client
0.03
1.01%
Computation
0.02
0.55%
Writing Output
0.05
1.65%
70000
34.81696
Chapter 9
The HPSAMPLE Procedure

Contents
Overview: HPSAMPLE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
119
Getting Started: HPSAMPLE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . .
120
Syntax: HPSAMPLE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
121
PROC HPSAMPLE statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
121
CLASS statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
123
PERFORMANCE statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
123
TARGET statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
123
VAR statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
123
Details: HPSAMPLE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
124
Class Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
124
Displayed Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
124
ODS Table Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
125
Examples: HPSAMPLE Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
125
Example 9.1: Running PROC HPSAMPLE on the Client . . . . . . . . . . . . . . .
125
Example 9.2: Running with Client Data on the SAS Appliance . . . . . . . . . . . .
126
Example 9.3: Running with Data on the SAS Appliance . . . . . . . . . . . . . . . .
128
Overview: HPSAMPLE Procedure

The HPSAMPLE procedure is a high-performance procedure that performs either simple random sampling
or stratified sampling. The HPSAMPLE procedure runs in either single-machine mode or distributed mode.
The HPSAMPLE procedure creates the following:
one output data set, which contains the sample data set
one performance table, which contains performance information
one frequency table, which contains the frequency information for the population and sample
Because the HPSAMPLE procedure is a high-performance analytical procedure, it also does the following:
computations
120 F Chapter 9: The HPSAMPLE Procedure
exploits all the available cores and concurrent threads, regardless of execution mode (currently HPSAMPLE is only using a single thread)
Topics.
Getting Started: HPSAMPLE Procedure

The following example shows a 10% stratified sampling, with the target variable BAD used by the
HPSAMPLE procedure as a stratum:
proc hpsample data=Sampsio.Hmeq out=Smp samppct=10 seed=1234 partition;
var loan derog mortdue value yoj delinq
clage ninq clno debtinc;
class bad reason job;
target bad;
run;
proc print data=Smp;run;
The input data set Sampsio.Hmeq includes information about 5,960 fictitious mortgages. Each observation represents an applicant for a home equity loan, and all applicants have an existing mortgage. The
SAMPPCT=10 option specifies that 10% of the input data be sampled. The SEED option specifies that the
random seed used in the sampling process be 1234. The PARTITION option specifies that the output data set,
Smp, include an indicator that shows whether each observation is selected to the sample (1) or not (0). The
VAR statement specifies 10 numeric input variables, and the CLASS statement specifies three classification
input variables. All these variables are included in the output sample. The binary TARGET variable BAD
indicates whether an applicant eventually defaulted or was ever seriously delinquent. The TARGET statement
triggers stratified sampling, which enables you to sample each subpopulation in the target variable (stratum)
independently. The displayed output contains a performance table (Figure 9.1) that shows the performance
environment information and a frequency table (Figure 9.2) that shows the frequency of observations in each
level of BAD.
Figure 9.1 Performance Information

Execution Mode
Single-Machine
Number of Threads 1
Figure 9.2 Frequency Table

One Target Stratified Sampling Frequency Table
Target Level Number of Obs Number of Samples
0
4771
478
1189
118
Syntax: HPSAMPLE Procedure F 121
Syntax: HPSAMPLE Procedure

The following statements are available in the HPSAMPLE procedure:
PROC HPSAMPLE < options > ;
VAR variable < variable ... variable > ;
CLASS variable < variable ... variable > ;
TARGET variable < variable ... variable > ;
Either a VAR or a CLASS statement is required for simple random sampling; both the CLASS and TARGET
statements are required for stratified sampling.
PROC HPSAMPLE statement

PROC HPSAMPLE < options > ;
The PROC HPSAMPLE statement invokes the procedure.

You can specify the following options:
DATA=< libref. >table
names the table (SAS data set or database table) that you want to sample from. The default is the most
recently opened or created data set. If the data are already distributed, the procedure reads the data
alongside the distributed database. See the section Single-machine Mode and Distributed Mode on
page 10 for the various execution modes and the section Alongside-the-Database Execution on page
15 for the alongside-the-database model.
NONORM
distinguishes target values that share the same normalized value when you perform stratified sampling
or oversampling. For example, if a target has three distinct values, A, B, and b, and you want to
treat B and b as different levels, you need to use NONORM. By default, B and b are treated as
the same level. PROC HPSAMPLE normalizes a value as follows:
1. Leading blanks are removed.
2. The value is truncated to 32 characters.
3. Letters are changed from lowercase to uppercase.
Note: In the oversampling case, there is no normalization for levels by default. If you do not specify
this option, you need to specify a normalized event value in the EVENT= option.
OUT=< libref. >SAS-data-set
names the SAS data set that you want to output the sample to. If you run alongside the database, you
need to specify a data set that has the same database libref as the input data set and make sure it does
not already exist in the database. This option is required.
Note: This SAS data set will contain the sample data set, which includes variables that are specified
in VAR and CLASS statements. If you also specify the PARTITION option, the output includes one
more column, _PartInd_. In the oversampling case, an additional column, _Freq_, is provided. It is
calculated as the ratio of rare levels proportion in the population to its proportion in the sample.
PARTITION
produces an output data set that has the same number of rows as the input data set but has an additional
partition indicator (_PARTIND_), which indicates whether an observation is selected to the sample (1)
or not (0). If you also specify the SAMPPCT2= option, _PARTIND_ indicates whether an observation
is selected to the sample1 (1), the sample2 (2), or the rest (0).
PARTINDNAME=partition-indicator-name
renames the partition indicator (_PARTIND_) to the specified partition-indicator-name.

SEED=random-seed
specifies the seed for the random number generator. If you do not specify random-seed or you specify
it as a nonpositive number, the seed is set to be the default 12345. The SEED option enables you to
reproduce the same sample output.
You can specify the following options only for simple random sampling and stratified sampling:
SAMPOBS=number
specifies the minimum number of observations you want to sample from the input data. The value of
number must be a positive integer. If number exceeds the total number of observations in the input
data, the output sample has the same number of observations as the input data set.
SAMPPCT=sample-percentage
specifies the sample percentage to be used by PROC HPSAMPLE. The value of sample-percentage
should be a positive number less than 100. For example, SAMPPCT=50.5 specifies that you want to
sample 50.5% of data.
N OTE : You must specify either the SAMPOBS or the SAMPPCT option if you want to perform simple
random sampling or stratified sampling. If you specify both options, only the SAMPPCT option is honored.
SAMPPCT2=sample-percentage
partitions the input data into three parts when specified along with the SAMPPCT= and PARTITION
options. The percentage of the sample whose _PARTIND_=1 is specified in the SAMPPCT= option,
the percentage of the sample whose _PARTIND_=2 is specified in the SAMPPCT2= option, and the
percentage of the sample whose _PARTIND_=0 is 100 minus the sum of the values of the SAMPPCT=
and SAMPPCT2= options. The sum of the sample-percentages specified in this option and in the
SAMPPCT2= option must be a positive number less than 100.
You can specify the following options only for oversampling:
EVENT=rare-event-level
specifies the rare event level. If you specify this option, PROC HPSAMPLE uses an oversampling
technique to adjust the class distribution of a data set, and the following two options are required.
SAMPPCTEVT=sample-event-percentage
specifies the sample percentage from the event level. The value of sample-event-percentage should be
a positive number less than 100. For example, SAMPPCTEVT=50.5 specifies that you want to sample
50.5 percent of the rare event level.
CLASS statement F 123
EVENTPROP=event-proportion
specifies the proportion of rare events in the sample. The value of event-proportion should be a positive
number less than 1. For example, EVENTPROP=0.3 specifies that you want the ratio between rare
events and not rare events to be 3:7.
CLASS statement
CLASS variable < variable ... variable > ;
The CLASS statement specifies one or more classification variables to be included in the sample. At least one
variable is required. A variable can be character or numeric. The CLASS and VAR statements are mutually
exclusive.
N OTE : Each variable in the TARGET statement must be specified in the CLASS statement. And the order of
Target variables in the output frequency table is the same as the order of these varibles specified in CLASS
statement.
PERFORMANCE statement
The PERFORMANCE statement defines performance parameters for multithreaded and distributed computing, passes variables about the distributed computing environment, and requests detailed results about the
performance characteristics of a SAS high-performance analytical procedure.
You can also use the PERFORMANCE statement to control whether a SAS high-performance analytical
procedure executes in single-machine mode or distributed mode.
The PERFORMANCE statement for SAS high-performance analytical procedures is documented in the
section PERFORMANCE Statement on page 36 of Chapter 3, Shared Concepts and Topics.
N OTE : PROC HPSAMPLE does not support multithreading in this release.
TARGET statement
TARGET variable < variable ... variable > ;
The TARGET statement specifies classification variables to be used for stratification. Each variable must be
specified in the CLASS statement. Currently, up to four target variables are supported for stratified sampling,
and one target variable is supported for oversampling. The maximum number of levels (that is, distinct
values) in any target variable is 256.
VAR statement
VAR variable < variable ... variable > ;
The VAR statement specifies one or more numeric variables to be included in the sample. At least one
variable is required; all variables must be numeric. You can use this statement to include only the variables
of interests in your sample. The CLASS and VAR statements are mutually exclusive.
Details: HPSAMPLE Procedure

Class Level
For classification variables, a level is an observed value that is distinct after formatting, removal of beginning
and ending white space, and capitalization. For example, the values MyLevel and MYLEVEL are treated as
a single level in the data set. Class variables can be numeric, and the same levelization rules apply. For
example, 3.000002 and 3.0000001 are treated as the same level if they are formatted using BEST3.
Displayed Output
The following sections describe the output that PROC HPSAMPLE produces by default. The output is
organized into various tables, which are discussed in the order of their appearance.
The Performance Information table is produced by default. It displays information about the execution
mode. For single-machine mode, the table displays the number of threads used. For distributed mode, the
table displays the grid mode (symmetric or asymmetric), the number of compute nodes, and the number of
threads per node.
If you specify the DETAILS option in the PERFORMANCE statement, the procedure also produces a
Timing table in which elapsed times (absolute and relative) for the main tasks of the procedure are
displayed.
Timing Table
The Timing Table lists the timing information for various computational stages of the procedure.
Frequency Information Table

For simple random sampling, the Frequency Information Table lists the number of observations in the
input data set and in the sample output data set.
For stratified sampling, the Frequency Information Table table lists the respective frequency in each stratum
for the input data and the sample. If one target variable is specified, each level of the target variable represents
a stratum; if two target variables are specified, a combination of the levels of two target variables represents a
stratum.
Example 9.1: Running PROC HPSAMPLE on the Client F 125
ODS Table Names

Each table that the HPSAMPLE procedure creates has a name associated with it. You must refer to the table
by this name when you use ODS statements. These names are listed in Table 9.1.
Table 9.1 ODS Tables Produced by PROC HPSAMPLE
Table Name
Description
Required Statement / Option
PerformanceInfo
Information about the high-performance computing

environment
Timing information for various computational stages
of the procedure
Frequency table of input data set and output sample
(when target variables are used, this table contains
stratification information for population and sample)
Default output
Timing
FreqTable
DETAILS (PERFORMANCE
statement)
Default output
Examples: HPSAMPLE Procedure

Example 9.1: Running PROC HPSAMPLE on the Client
This example demonstrates how to use PROC HPSAMPLE to perform simple random sampling on the
Sampsio.Hmeq data set, which resides on the client.
following statements, the client performs all computations:
/*sampsio is a libref for a data source on the client.*/
proc hpsample data=sampsio.hmeq out=out1 sampobs=20 seed=13579;
class job reason;
var loan value delinq derog;
run;
proc print data=out1;
run;
Output 9.1.1 shows the performance environment information.


Execution Mode
Single-Machine
Number of Threads 1
Output 9.1.2 shows the number of observations in the data set sampsio.hmeq and the number of samples.
Output 9.1.2 Frequency Table
Simple Random Sampling
Frequency Table
Number of Obs Number of Samples
5960
22
Output 9.1.3 shows the sample data.

Output 9.1.3 Sample Data
Obs JOB
REASON LOAN VALUE DELINQ DEROG
1 Other
HomeImp
4900
65774
2 Other
HomeImp
5700
82923
3 Other
HomeImp
7000 124827
DebtCon
8300
75081
5 Office
DebtCon 10000 125500
6 Other
11100
61406
7 Other
HomeImp 11500
64037
8 ProfExe HomeImp 11900 105454
9 Mgr
HomeImp 15000 122400
10 Other
HomeImp 15000 107207
11 ProfExe HomeImp 15600 106824
12 Office
DebtCon 17000
69000
13 ProfExe DebtCon 17300
49100
14 Other
DebtCon 18000
60000
15 Office
DebtCon 21100
98000
16 Other
HomeImp 21400 103427
17 Mgr
HomeImp 22400 121601
18 Other
HomeImp 25000 202500
19 Other
DebtCon 25500
43031
20 Mgr
DebtCon 27500 149877
21 ProfExe DebtCon 36500 195729
22 Self
DebtCon 70300 294169
Example 9.2: Running with Client Data on the SAS Appliance

This example uses the same data set as is used in Example 9.1. This example demonstrates how to use PROC
HPSAMPLE to perform stratified sampling.
When the input data set resides on the client and a PERFORMANCE statement with a NODES= option is
specified, as in the following statements, PROC HPSAMPLE copies the data set to the SAS appliance, where
the sampling is performed:
Example 9.2: Running with Client Data on the SAS Appliance F 127

option set=GRIDHOST="&GRIDHOST";
option set=GRIDINSTALLLOC="&GRIDINSTALLLOC";
proc hpsample data=sampsio.hmeq out=out2 samppct=10 seed=13579 partition;
class job reason;
target job;
performance nodes = 2;
run;
proc print data=out2(obs=15);
run;


Host Node
bigmath.unx.sas.com
Execution Mode
Distributed
Output 9.2.2 shows the frequency information for each level of target variable JOB in the data set Sampsio.Hmeq and in the sample.
One Target Stratified Sampling Frequency Table
279
28
MGR
767
77
OFFICE
948
95
OTHER
2388
239
PROFEXE
1276
127
SALES
109
11
SELF
193
19
Output 9.2.3 shows the first 15 output sample observations that contain _PARTIND_, which indicates
whether the observation is selected for the sample (1) or not (0).
Output 9.2.3 Sample Output with Partition Indicator

Obs JOB
REASON LOAN VALUE DELINQ DEROG _PartInd_
1 Other HomeImp
1100
39025
2 Other HomeImp
1300
68400
3 Other HomeImp
1500
16700
1500
5 Office HomeImp
1700 112000
6 Other HomeImp
1700
40320
7 Other HomeImp
1800
57037
8 Other HomeImp
1800
43034
9 Other HomeImp
2000
46740
10 Sales HomeImp
2000
62250
11
2000
12 Office HomeImp
2000
29800
13 Other HomeImp
2000
55000
14 Mgr
2000
87400
15 Other HomeImp
2100
83850
Example 9.3: Running with Data on the SAS Appliance

This example uses the same data set as in Example 9.1. It demonstrates how to use PROC HPSAMPLE to
perform oversampling.
When the input data set resides on the SAS appliance, the SAS appliance performs all samplings, writes
out a sample on the SAS appliance, and reports the frequency results back to the client. In the following
statements, the input data resides in the MyLib library (which is a distributed data source), and the output
data set is a distributed data set in the MyLib library. You can use PROC DATASETS to delete the output
table if it already exists on the SAS appliance. The ods output FreqTable=Freqtab; statement saves
the frequency table to a SAS data set called Freqtab on the client.
/*MyLib is a libref for a distributed data source
In this case, the computation is automatically done
on the SAS Appliance.*/
option set=GRIDHOST
= "&GRIDHOST";
option set=GRIDMODE = "&GRIDMODE";
libname MyLib &LIBTYPE
server ="&GRIDDATASERVER"
user
=&USER
password=&PASSWORD
database=&DATABASE;
proc delete data=MyLib.out_smp_ex3; run;
proc hpsample data=MyLib.hmeq out=MyLib.out_smp_ex3 seed=13579 partition
samppctevt=80 eventprop=.2 event="SALES";
class job;
target job;
Example 9.3: Running with Data on the SAS Appliance F 129
ods output FreqTable=Freqtab;

run;


Host Node
greenarrow.unx.sas.com
Execution Mode
Distributed
16
Output 9.3.2 shows the number of observations in each level of target variable JOB in the data set MyLib.Hmeq
and in the sample. After oversampling, the proportion of SALES level is adjusted to 20% in the sample from
the original 1.8% in the population.
Oversampling Frequency Table
MGR
279
15
767
47
OFFICE
948
56
OTHER
2388
142
PROFEXE
1276
77
SALES
109
87
SELF
193
15
130
Chapter 10
The HPSUMMARY Procedure

Contents
Overview: HPSUMMARY Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
131
PROC HPSUMMARY Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
132
133
PROC HPSUMMARY Contrasted with Other Procedures . . . . . . . . . . . . . . .
133
Getting Started: HPSUMMARY Procedure . . . . . . . . . . . . . . . . . . . . . . . . . .
133
Syntax: HPSUMMARY Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
135
PROC HPSUMMARY Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . .
136
Statistic Keywords . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
139
CLASS Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
140
FREQ Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
141
OUTPUT Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
141
143
TYPES Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
143
VAR Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
144
WAYS Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
145
WEIGHT Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
146
Details: HPSUMMARY Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
147
How PROC HPSUMMARY Groups Data . . . . . . . . . . . . . . . . . . . . . . . .
147
Computational Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
148
Statistical Computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
149
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Keywords and Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
151
153
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
160
Overview: HPSUMMARY Procedure

The HPSUMMARY procedure computes basic descriptive statistics for variables in a SAS data set. It is a
high-performance version of the SUMMARY procedure in Base SAS. PROC HPSUMMARY runs in either
single-machine mode or distributed mode.
When run in distributed mode, the HPSUMMARY procedure enables you to summarize data that have been
distributed to the grid for parallel execution. The output data that PROC HPSUMMARY creates can then be
written in parallel back to the grid data store.
132 F Chapter 10: The HPSUMMARY Procedure
In distributed mode, it is recommended that input data reside on the grid and that results be output back to
the grid. Although it is possible to use PROC HPSUMMARY on data that do not reside on the grid or to
produce result tables that do not reside on the grid, this usage is not recommended because of the overhead of
transferring data to and from the grid.
PROC HPSUMMARY provides functionality similar to that of the SUMMARY procedure in Base SAS. Its
syntax, options, and underlying concepts are also similar. Because of this similarity, documentation for the
SUMMARY procedure can be useful in understanding PROC HPSUMMARY. For more information about
the SUMMARY procedure, see the Base SAS Procedures Guide.
PROC HPSUMMARY Features

PROC HPSUMMARY provides data summarization tools to compute descriptive statistics for variables
across all observations and within groups of observations. For example, PROC HPSUMMARY does the
following:
calculates descriptive statistics based on moments
calculates and estimates quantiles, which includes the median
calculates confidence limits for the mean
identifies extreme values
performs a t test
PROC HPSUMMARY does not display output. You can use the OUTPUT statement to store the statistics in
a SAS data set.
PROC HPSUMMARY provides a vehicle for the parallel execution of summarization in a distributed
computing environment. The following list summarizes the basic features of PROC HPSUMMARY:
provides the ability to execute summarization in parallel
enables you to control the level of parallelism per execution node and the number of nodes to engage
is highly multithreaded
needed
Because the HPSUMMARY procedure is a high-performance analytical procedure, it also does the following:
computations
Topics.
Single-Machine and Distributed Execution Modes F 133

The HPSUMMARY procedure enables you to perform analyses either on a single computer (single-machine
mode) or on multiple computers that are connected in a grid configuration (distributed mode). For more
information about these execution modes, see the section Processing Modes on page 10 in Chapter 3,
Shared Concepts and Topics.
In single-machine mode, you can take advantage of multiple processors and cores in a single machine, and
you can control the number of parallel threads.
In distributed mode, you can take advantage of the collective processing resources of multiple machines. You
can control both the number of parallel threads per execution node and the number of nodes to engage. One
or more copies of the summarization code are executed in parallel on each node. You can read data in parallel
from and write data in parallel to a supported database management system (DBMS) on each node in the
grid, thus greatly reducing processing time for large volumes of data. The distributed mode of execution has
two variations:
In the client-data (local-data) model of distributed execution, the input data are not stored on the grid
computing appliance but are distributed to it from the client during execution of the HPSUMMARY
procedure.
appliance. The data are stored in the distributed database, and the summarization code that runs on
each node can read and write the data in parallel during execution of the procedure. Instead of being
moved across the network and possibly back to the client machine, data are passed locally between the
processes on each node of the appliance. In general, especially with large data sets, the best PROC
HPSUMMARY performance can be achieved if execution is alongside the database.
PROC HPSUMMARY Contrasted with Other Procedures

By default, PROC SUMMARY generates all CLASS variable combination types and requires the NWAY
option to generate only the n-way. By default, PROC HPSUMMARY generates only the n-way, and requires
the ALLTYPES option to generate all of the types.
Getting Started: HPSUMMARY Procedure

This example illustrates a simple use of the HPSUMMARY procedure to summarize using the grid in
distributed mode. The following DATA step creates a data set that consists of test scores:
data gridlib.grades;
input Name $ 1-8 Gender $ 11 Status $13 Year $ 15-16
Section $ 18 Score 20-21 FinalGrade 23-24;
datalines;
Abbott
F 2 97 A 90 87
Branford
Crandell
Dennison
Edgar
Faust
Greeley
Hart
Isley
Jasper
;
run;
M
M
M
F
M
F
F
M
M
1
2
1
1
1
2
1
2
1
98
98
97
98
97
97
98
97
97
A
B
A
B
B
A
B
A
B
92
81
85
89
78
82
84
88
91
97
71
72
80
73
91
80
86
93
The following statements read this data set and analyze the data for the two-way combination of CLASS
variables and across all observations. To run these statements successfully, you need to set the macro variables
GRIDHOST and GRIDINSTALLLOC to resolve to appropriate values, or you can replace the references to the
macro variables in the example with the appropriate values.
proc hpsummary data=gridlib.grades;
var Score;
class Status Year;
types () status*year;
output out=gridlib.result;
run;
proc print data=gridlib.result;
run;
Figure 10.1 displays the tables produced by the HPSUMMARY procedure. The Performance Information
table shows that PROC HPSUMMARY ran in distributed mode and the Data Access Information table
shows that the date were accessed in parallel symmetric mode.
Figure 10.1 HPSUMMARY Output
The HPSUMMARY Procedure

Host Node
greenarrow.unx.sas.com
Execution Mode
Distributed
16

Data
Engine
Role
GRIDLIB.grades GREENPLM Input

GRIDLIB.result
Path
Parallel, Symmetric
GREENPLM Output Parallel, Symmetric
Syntax: HPSUMMARY Procedure F 135

Obs status year _type_ _freq_ _stat_
1 1
98
score
3 MIN
10 STD
84.0000
4.7140
10 MAX
92.0000
10 MIN
78.0000
5 1
98
3 MAX
92.0000
6 2
98
1 MIN
81.0000
7 2
97
3 N
3.0000
8 1
97
3 STD
6.5064
9 1
98
3 N
10 2
97
3 MAX
11
10 N
3.0000
90.0000
10.0000
12 2
98
1 MEAN 81.0000
13 2
98
1 MAX
81.0000
14 2
97
3 MIN
82.0000
15 2
98
1 STD
16 1
97
3 N
3.0000
17 2
97
3 STD
4.1633
18 1
97
3 MEAN 84.6667
19 2
97
3 MEAN 86.6667
20 1
97
3 MAX
91.0000
21 1
97
3 MIN
78.0000
22
10 MEAN 86.0000
23 1
98
3 STD
24 1
98
3 MEAN 88.3333
4.0415
25 2
98
1 N
1.0000
Syntax: HPSUMMARY Procedure

The following statements are available in the HPSUMMARY procedure:
PROC HPSUMMARY < options > < statistic-keywords > ;
CLASS variables < / options > ;
FREQ variable ;
OUTPUT < OUT=SAS-data-set > < output-statistic-specifications > < / AUTONAME > ;
TYPES requests ;
VAR variables < / WEIGHT=weight-variable > ;
WAYS list ;
WEIGHT variable ;
You can also use the ATTRIB, FORMAT, LABEL, and WHERE statements and any global statements. For
more information, see SAS Statements: Reference.
PROC HPSUMMARY Statement

PROC HPSUMMARY < options > < statistic-keywords > ;
The PROC HPSUMMARY statement invokes the procedure. The HPSUMMARY procedure computes
descriptive statistics for variables across all observations or within groups of observations.
Table 10.1 summarizes the available options in the PROC HPSUMMARY statement by function. The
options are then described fully in alphabetical order in the section Optional Arguments on page 136. For
information about the statistic-keywords, see the section Statistic Keywords on page 139.
Table 10.1 PROC HPSUMMARY Statement Options
Option
Description
Basic Options
DATA=
PCTLDEF=

Specifies the mathematical definition used to compute quantiles
Option Related to Classification Level

MISSING
Uses missing values as valid values to create combinations of classification variables
Options Related to the Output Data Set
ALLTYPES
Computes statistics for all combinations of classification variables (not just the n-way)
CHARTYPE
Specifies that the _TYPE_ variable contain character values
Options Related to Statistical Analysis
ALPHA=
Specifies the confidence level for the confidence limits
EXCLNPWGT
Excludes observations with nonpositive weights from the analysis
QMARKERS=
Specifies the sample size to use for the P2 quantile estimation method
QMETHOD=
Specifies the quantile estimation method
QNTLDEF=
Specifies the mathematical definition used to compute quantiles
statistic-keywords Selects the statistics
VARDEF=
Specifies the variance divisor
Optional Arguments
You can specify the following options in the PROC HPSUMMARY statement:
ALLTYPES
ALLWAYS
requests that PROC HPSUMMARY compute descriptive statistics for all combinations of classification
variables. By default, PROC HPSUMMARY generates only the n-way. For more information, see the
section How PROC HPSUMMARY Groups Data on page 147.
ALPHA=value
specifies the confidence level to compute the confidence limits for the mean. The percentage for the
confidence limits is 100(1value). For example, ALPHA=0.05 results in a 95% confidence limit. You
PROC HPSUMMARY Statement F 137
can specify any value between 0 and 1. The default is 0.05. To compute confidence limits, specify the
statistic-keyword CLM, LCLM, or UCLM. See the section Confidence Limits on page 149.
CHARTYPE
specifies that the _TYPE_ variable in the output data set is a character representation of the binary
value of _TYPE_. The length of the variable equals the number of classification variables. When you
specify more than 32 classification variables, _TYPE_ automatically becomes a character variable. See
the section Output Data Set on page 152.
DATA=SAS-data-set
names the SAS data set to be used as the input data set. The default is the most recently created data
set.
EXCLNPWGT
EXCLNPWGTS
excludes observations with nonpositive weight values (0 or negative) from the analysis. By default,
PROC HPSUMMARY treats observations with negative weights like observations with zero weights
and counts them in the total number of observations. See the WEIGHT= option and the section
WEIGHT Statement on page 146.
MISSING
considers missing values as valid values to create the combinations of classification variables. Special
missing values that represent numeric valuesthe letters A through Z and the underscore (_) character
are each considered as a separate value. If you omit MISSING, then PROC HPSUMMARY excludes
the observations with a missing classification variable value from the analysis. See SAS Language
Reference: Concepts for a discussion of missing values that have special meanings.
PCTLDEF=1 | 2 | 3 | 4 | 5
is an alias for the QNTLDEF= option.

QMARKERS=number
specifies the default number of markers to use for the P 2 quantile estimation method. The number of
markers controls the size of fixed memory space.
The value of number must be an odd integer greater than 3. The default value depends on which
quantiles you request. For the median (P50), number is 7. For the quantiles (P25 and P50), number
is 25. For the quantiles P1, P5, P10, P75 P90, P95, or P99, number is 105. If you request several
quantiles, then PROC HPSUMMARY uses the largest value of number .
You can improve the accuracy of the estimate by increasing the number of markers above the default
settings; you can conserve memory and computing time by reducing the number of markers. See the
section Quantiles on page 150.
QMETHOD=OS | P2
specifies the method that PROC HPSUMMARY uses to process the input data when it computes quantiles. If the number of observations is less than or equal to the QMARKERS= value and QNTLDEF=5,
then both methods produce the same results. The QMETHOD= option can take either of the following
values:
OS
specifies that PROC HPSUMMARY use order statistics.

N OTE : This technique can be very memory-intensive.
P2
specifies that PROC HPSUMMARY use the P 2 method to approximate the quantile. When
QMETHOD=P2, PROC HPSUMMARY does not compute MODE or weighted quantiles. In
addition, reliable estimations of some quantiles (P1, P5, P95, P99) might not be possible for
some data sets.
The default is OS. See the section Quantiles on page 150 for more information.
QNTLDEF=1 | 2 | 3 | 4 | 5
PCTLDEF=1 | 2 | 3 | 4 | 5
specifies the mathematical definition that PROC HPSUMMARY uses to calculate quantiles when
QMETHOD=OS. The default is 5. To use QMETHOD=P2, you must use QNTLDEF=5. See the
section Quantile and Related Statistics on page 157.
VARDEF=divisor
specifies the divisor to use in the calculation of the variance and standard deviation. Table 10.2 shows
the possible values for divisor and their associated formulas.
Table 10.2 Values for VARDEF= Option
divisor
DF
N
WDF
WEIGHT | WGT
Description
Degrees of freedom
Sum of weights minus one
Sum of weights
Formula for Divisor

n 1
nP

w
1
i
i
P
i wi
The procedure computes the variance as CSS divided by divisor , where the corrected sum of squares
CSS is defined by the following formula:
X
.xi x/
CSS D
N 2
When you weight the analysis variables, the formula for CSS is
X
CSS D
wi .xi xN w /2
where xN w is the weighted mean.
The default is DF. To compute the standard error of the mean, confidence limits for the mean, or the
Students t-test, you must use this default value.
When you use the WEIGHT statement and VARDEF=DF, the variance is an estimate of 2 , where
the variance of the ith observation is var.xi / D 2 =wi and wi is the weight for the ith observation.
This method yields an estimate of the variance of an observation with unit weight. When you use
the WEIGHT statement and VARDEF=WGT, the computed variance is asymptotically (for large n)
an estimate of 2 =w,
N where wN is the average weight. This method yields an asymptotic estimate of
the variance of an observation with average weight. See the section Keywords and Formulas on
page 153.
Statistic Keywords F 139
Statistic Keywords
Optional statistic-keywords specify which statistics to compute and the order to display them in the output.
Table 10.3 lists the keywords that are available in the PROC HPSUMMARY statement. The definitions of
the keywords and the formulas for the associated statistics are listed in the section Keywords and Formulas
on page 153.
Table 10.3 Statistic Keywords in the PROC HPSUMMARY
Statement
Descriptive Statistic Keywords:

CLM
CSS
CV
KURTOSIS | KURT
LCLM
MAX
MEAN
MIN
MODE
N
NMISS
RANGE
SKEWNESS | SKEW
STDDEV | STD
STDERR
SUM
SUMWGT
UCLM
USS
VAR
Quantile Statistic Keywords:

MEDIAN | P50
P1
P5
P10
P20
P40
P70
Q1 | P25
Q3 | P75
P90
P95
P99
P30
P60
P80
QRANGE
Hypothesis Testing Keywords:

PROBT | PRT
The default values are N, MEAN, STD, MIN, and MAX. To compute standard error, confidence limits for
the mean, and the Students t-test, you must use the default value of the VARDEF= option, which is DF. To
compute skewness or kurtosis, you must use VARDEF=N or VARDEF=DF. Use CLM or both LCLM and
UCLM to compute a two-sided confidence limit for the mean. Use only LCLM or UCLM to compute a
one-sided confidence limit. The definitions of the keywords and the formulas for the associated statistics are
listed in the section Keywords and Formulas on page 153.
CLASS Statement
CLASS variables < / options > ;
The CLASS statement names the classification variables to be used as explanatory variables in the analysis.
These variables enter the analysis not through their values, but through levels to which the unique values
are mapped. For more information, see the section Levelization of Classification Variables (Chapter 4,
SAS/STAT Users Guide: High-Performance Procedures).
Levels of classification variables are ordered by their external formatted values, except for numeric variables
with no explicit format, which are ordered by their unformatted (internal) values.
Required Argument
variables
specifies one or more variables that the procedure uses to group the data. Variables in a CLASS
statement are referred to as classification variables. Classification variables are numeric or character.
Classification variables can have continuous values, but they typically have a few discrete values that
define levels of the variable. You do not have to sort the data by classification variables.
Use the TYPES statement or the WAYS statement to control which classification variables PROC
HPSUMMARY uses to group the data. See the section How PROC HPSUMMARY Groups Data on
page 147.
To reduce the number of classification variable levels, use a FORMAT statement to combine variable
values. When a format combines several internal values into one formatted value, PROC HPSUMMARY outputs the lowest internal value.
Optional Arguments
GROUPINTERNAL
specifies that formats are not to be applied to the classification variables when PROC HPSUMMARY
groups the values to create combinations of classification variables. This option saves computer
resources when the numeric classification variables contain discrete values. See the section Computational Resources on page 148.
MISSING
considers missing values as valid values for the classification variable levels. Special missing values
that represent numeric valuesthe letters A through Z and the underscore (_) characterare each
considered as a separate value. If you omit the MISSING option, then PROC HPSUMMARY excludes
the observations with a missing classification variable value from the analysis.
By default, if an observation contains a missing value for any classification variable, then PROC
HPSUMMARY excludes that observation from the analysis. If you specify the MISSING option in the
PROC HPSUMMARY statement, then the procedure considers missing values as valid levels for the
combination of classification variables.
Specifying the MISSING option in the CLASS statement enables you to control the acceptance of
missing values for individual classification variables.
See SAS Language Reference: Concepts for a discussion of missing values that have special meaning.
FREQ Statement F 141
FREQ Statement
FREQ variable ;
The FREQ statement specifies a numeric variable that contains the frequency of each observation.
Required Argument
variable
specifies a numeric variable whose value represents the frequency of the observation. If you use the
FREQ statement, then the procedure assumes that each observation represents n observations, where n
is the value of variable. If n is not an integer, then SAS truncates it. If n is less than 1 or is missing,
then the procedure does not use that observation to calculate statistics.
The sum of the frequency variable represents the total number of observations.
OUTPUT Statement
OUTPUT < OUT=SAS-data-set > < output-statistic-specifications > < / AUTONAME > ;
The OUTPUT statement writes statistics to a new SAS data set. You can use multiple OUTPUT statements
to create several OUT= data sets.
Optional Arguments
OUT=SAS-data-set
names the new output data set. If SAS-data-set does not exist, then PROC HPSUMMARY creates it.
If you omit the OUT= option, then the data set is named DATAn, where n is the smallest integer that
makes the name unique.
output-statistic-specifications
specifies the statistics to store in the OUT= data set and names one or more variables that contain the
statistics. The form of the output-statistic-specification is
statistic-keyword < (variable-list) >=< names >
where
statistic-keyword
specifies which statistic to store in the output data set. Table 10.4 lists the statistic-keywords
that are available in the OUTPUT statement.
Table 10.4 Statistics Keywords in the OUTPUT Statement
Descriptive Statistic Keywords:

CSS
CV
KURTOSIS | KURT
LCLM
RANGE
SKEWNESS | SKEW
STDDEV | STD
STDERR
Table 10.4 (continued)
MAX
MEAN
MIN
MODE
N
NMISS
SUM
SUMWGT
UCLM
USS
VAR
Quantile Statistic Keywords:

MEDIAN | P50
P1
P5
P10
P20
P40
P70
Q1 | P25
Q3 | P75
P90
P95
P99
P30
P60
P80
QRANGE
Hypothesis Testing Keywords:

PROBT | PRT
By default the statistics in the output data set automatically inherit the analysis variables format,
informat, and label. However, statistics computed for N, NMISS, SUMWGT, USS, CSS, VAR,
CV, T, PROBT, PRT, SKEWNESS, and KURTOSIS do not inherit the analysis variables format
because this format might be invalid for these statistics (for example, dollar or datetime formats).
If you omit a variable-list and names, then PROC HPSUMMARY allows the statistic-keyword
only once in a single OUTPUT statement, unless you also use the AUTONAME option.
The definitions of the keywords and the formulas for the associated statistics are listed in the
section Keywords and Formulas on page 153.
variable-list
specifies the names of one or more numeric analysis variables whose statistics you want to store
in the output data set. By default, statistics are stored for all numeric analysis variables.
names
specifies one or more names for the variables in output data set to contain the analysis variable
statistics. The first name contains the statistic for the first analysis variable; the second name
contains the statistic for the second analysis variable; and so on. The default value is the analysis
variable name. If you specify the AUTONAME option, then the default is the combination
of the analysis variable name and the statistic-keyword . If you use the CLASS statement
and an OUTPUT statement without an output-statistic-specification, then the output data set
contains five observations for each combination of classification variables: the value of N, MIN,
MAX, MEAN, and STD. If you use the WEIGHT statement or the WEIGHT option in the
VAR statement, then the output data set also contains an observation with the sum of weights
(SUMWGT) for each combination of classification variables.
If you specify variable-list , then PROC HPSUMMARY uses the order in which you specify
the analysis variables to store the statistics in the output data set variables. You can use the
AUTONAME option to request that PROC HPSUMMARY generate unique names for multiple
variables and statistics.
AUTONAME
requests that PROC HPSUMMARY create a unique variable name for an output statistic when
you do not assign the variable name in the OUTPUT statement. This action is accomplished
by appending the statistic-keyword to the input variable name. For example, the following
statement produces the x_Min variable in the output data set:
output min(x)=/autoname;
AUTONAME activates the SAS internal mechanism that automatically resolves conflicts in the
variable names in the output data set so that duplicate variables do not generate errors. As a
result, the following statement produces two variables, x_Min and x_Min2, in the output data set:
output min(x)= min(x)=/autoname;
the performance characteristics of the HPSUMMARY procedure.
You can also use the PERFORMANCE statement to control whether PROC HPSUMMARY executes in
It is important to remember the distinction between the NODES= and NTHREADS= options. The NODES=
option specifies the number of separate grid nodes that participate in the PROC HPSUMMARY execution,
and the NTHREADS= option influences how many threads are used by the HPSUMMARY procedure
instance that runs on each node. If the data are located on the grid, then all nodes must be engaged; therefore,
the NODES= option might be overridden. Specifying NODES=0 causes PROC HPSUMMARY to execute in
single-machine mode only. Setting the NTHREADS= option to a value that is greater than the CPU count on
each grid node is not likely to improve overall throughput.
TYPES Statement
TYPES requests ;
The TYPES statement identifies which of the possible combinations of classification variables to generate.
The TYPES statement requires the specification of a CLASS statement.
Required Argument
requests
specifies which of the 2k combinations of classification variables PROC HPSUMMARY uses to create
the types, where k is the number of classification variables. A request includes one classification
variable name, several classification variable names separated by asterisks, or ().
To request classification variable combinations quickly, use a grouping syntax by placing parentheses
around several variables and joining other variables or variable combinations. The examples in
Table 10.5 illustrate grouping syntax:
Table 10.5 Examples of Grouping Syntax
Request
Equivalent To
types A*(B C);

types (A B)*(C D);
types (A B C)*D;
types A*B A*C;

types A*C A*D B*C B*D;
types A*D B*D C*D;
You can use parentheses () to request the overall total (_TYPE_=0). If you do not need all types in the
output data set, then use the TYPES statement to specify particular subtypes rather than applying a
WHERE clause to the data set. Doing so saves time and computer memory.
Order of Analyses in the Output

The SUMMARY procedure writes analyses to the output in order of increasing values of the _TYPE_
variable. When PROC HPSUMMARY executes on the grid, the order of observations within the output is
not deterministic because the output is returned in parallel. You can sort the output as follows:
If output is directed back to the client, then to achieve an output order that is similar to the output of
PROC SUMMARY, you need to subsequently sort the data by _TYPE_ and the classification variables.
If output is directed back to the grid (so that the results are distributed), then there is no order within
the output. To retrieve the observations in order, you can execute an SQL query, specifying that the
selecting rows be returned in order by _TYPE_ and the classification variables.
The _TYPE_ variable is calculated even if no output data set is requested. For more information about the
_TYPE_ variable, see the section Output Data Set on page 152.
VAR Statement
VAR variables < / WEIGHT=weight-variable > ;
The VAR statement identifies the analysis variables and their order in the output. If you omit the VAR
statement, then PROC HPSUMMARY analyzes all numeric variables that are not listed in the other statements.
When all variables are character variables, PROC SUMMARY produces a simple count of observations. You
can use multiple VAR statements.
WAYS Statement F 145
Required Argument
variables
identifies one or more analysis variables and specifies their order in the results.
Optional Argument
WEIGHT=weight-variable
specifies a numeric variable whose values weight the values of the variables. The weight-variable does
not have to be an integer. Table 10.6 describes how PROC HPSUMMARY treats various values of the
weight-variable.
Table 10.6 Responses to Values of weight-variable
Value
PROC HPSUMMARY Response
Counts the observation in the total number of observations

Converts the value to zero and counts the observation in the total number of observations
Excludes the observation
Less than 0
Missing
To exclude observations that contain negative and zero weights from the analysis, use the EXCLNPWGT option in the PROC HPSUMMARY statement.
The weight-variable does not change how the procedure determines the range, extreme values, or
number of missing values.
To compute weighted quantiles, use QMETHOD=OS in the PROC HPSUMMARY statement. Skewness and kurtosis are not available with the WEIGHT= option.
When you use the WEIGHT= option, consider which value of the VARDEF= option is appropriate.
Use the WEIGHT= option in multiple VAR statements to specify different weights for the analysis
variables.
WAYS Statement
WAYS list ;
The WAYS statement specifies the number of ways to make unique combinations of classification variables.
You can use the TYPES statement to specify additional combinations of classification variables.
Required Argument
list
specifies one or more integers that define the number of classification variables to combine to form all
the unique combinations of classification variables. For example, you can specify 2 for all possible
pairs and 3 for all possible triples. The list can be specified in the following ways:
m
m1 m2 ... mn
m1,m2 ,...,mn
m TO n <BY increment >
m1,m2 , TO m3 <BY increment >,m4
The range of list is from 0 to the maximum number of classification variables.
The following statements are an example of creating two-way types for the classification variables A,
B, and C:
class A B C ;
ways 2;
The WAYS statement in this example is equivalent to specifying A*B, A*C, and B*C in the TYPES
statement.
WEIGHT Statement
WEIGHT weight-variable ;
The WEIGHT statement specifies weights for observations in the statistical calculations.
Required Argument
weight-variable
specifies a numeric weight-variable whose values weight the values of the analysis variables. The
values of weight-variable do not have to be integers. Table 10.7 describes how PROC HPSUMMARY
treats various values of weight-variable.
Table 10.7 Responses to Values of weight-variable
Value
PROC HPSUMMARY Response
Counts the observation in the total number of observations

Converts the value to zero and counts the observation in the total number of observations
Excludes the observation
Less than 0
Missing
Details: HPSUMMARY Procedure F 147
To exclude observations that contain negative and zero weights from the analysis, use the EXCLNPWGT option in the PROC HPSUMMARY statement.
C AUTION : Single extreme weight values can cause inaccurate results. When one (and only one)
weight value is many orders of magnitude larger than the other weight values (for example, 49 weight
values of 1 and one weight value of 1 1014 ), certain statistics might not be within acceptable accuracy
limits. The affected statistics are based on the second moment (such as standard deviation, corrected
sum of squares, variance, and standard error of the mean). Under certain circumstances, no warning is
written to the SAS log.
To compute weighted quantiles, you must use QMETHOD=OS in the PROC statement. Skewness and
kurtosis are not available with the WEIGHT statement.
PROC HPSUMMARY does not compute MODE when a weight variable is active. Instead, you can try
using the UNIVARIATE procedure when MODE needs to be computed and a weight variable is active.
If you use the WEIGHT= option in a VAR statement to specify a weight variable, then PROC
HPSUMMARY uses this variable instead to weight those VAR statement variables.
When you use the WEIGHT statement, consider which value of the VARDEF= option is appropriate.
See the section Keywords and Formulas on page 153 for more information.
Details: HPSUMMARY Procedure

How PROC HPSUMMARY Groups Data
Groups of observations are defined by specifying certain variables as classification variables in the CLASS
statement. Unique values of the n CLASS variables are used to partition the input data, and the resulting
summarized data (one observation per group) is called the n-way.
PROC HPSUMMARY can also combine the partitioned groups into larger groups by removing one or more
CLASS variables from consideration when grouping. There are 2n different groupings that can be generated
from n CLASS variables. Each of these groupings is a type, which appears in the output data set as a
variable named _TYPE_. Type 0 includes no CLASS variables and summarizes the entire input data set,
Type 1 includes only the last CLASS variable specified, and so on to Type 2n 1, which is the n-way.
By default, PROC HPSUMMARY generates only the n-way. The option ALLTYPES (or ALLWAYS) in the
PROC HPSUMMARY statement generates all 2n types. You can also use either of the following statements
to choose which types appear in the output data set:
The WAYS statement specifies how many CLASS variables appear in each output type. For example,
WAYS 1 produces types for each CLASS variable individually, WAYS 2 generates all n2 possible
pairs, and so on.
The TYPES statement explicitly specifies the desired types by CLASS variable name, such as TYPES
A A*B C (), where A*B might specify Type 6 and () specifies Type 0.
The TYPES statement controls which of the available classification variables PROC HPSUMMARY uses
to subgroup the data. The unique combinations of these active classification variable values that occur
together in any single observation of the input data set determine the data subgroups. Each subgroup that
PROC HPSUMMARY generates for a given type is called a level of that type. For all types, the inactive
classification variables can still affect the total observation count of the rejection of observations with missing
values. When you use a WAYS statement, PROC HPSUMMARY generates types that correspond to every
possible unique combination of n classification variables chosen from the complete set of classification
variables. For example
proc hpsummary;
class a b c d e;
ways 2 3;
output out=results;
run;
is equivalent to
proc hpsummary;
class a b c d e;
types a*b a*c a*d a*e b*c b*d b*e c*d c*e d*e
a*b*c a*b*d a*b*e a*c*d a*c*e a*d*e
b*c*d b*c*e c*d*e;
output out=results;
run;
If you omit the TYPES statement and the WAYS statement, then PROC HPSUMMARY uses all classification
variables to subgroup the data (the NWAY type) for the output data set.
Computational Resources
The total of unique classification values that PROC HPSUMMARY allows depends on the amount of
computer memory that is available. PROC HPSUMMARY uses the same memory allocation scheme across
all operating environments. When classification variables are involved, PROC HPSUMMARY must keep
a copy of each unique value of each classification variable in memory. You can estimate the memory
requirements to group the classification variable by calculating
Nc1 .Lc1 C K/ C Nc2 .Lc2 C K/ C : : : C Ncn .Lcn C K/
where Nci is the number of unique values for the classification variable, Lci is the combined unformatted
and formatted length of ci , and K is some constant on the order of 32 bytes (64 for 64-bit architectures).
When you use the GROUPINTERNAL option in the CLASS statement, Lci is simply the unformatted length
of ci .
The GROUPINTERNAL option can improve computer performance because the grouping process is based
on the internal values of the classification variables. If a numeric classification variable is not assigned a
format and you do not specify GROUPINTERNAL, then PROC HPSUMMARY uses the default format,
BEST12., to format numeric values as character strings. Then PROC HPSUMMARY groups these numeric
variables by their character values, which takes additional time and computer memory.
Each unique combination of classification variables c1i c2j for a given type forms a level in that type. See the
section TYPES Statement on page 143. You can estimate the maximum potential space requirements for
all levels of a given type, when all combinations actually exist in the data (a complete type), by calculating
W Nc1 Nc2 : : : Ncn
Statistical Computations F 149
where W is a constant based on the number of variables analyzed and the number of statistics calculated
(unless you request QMETHOD=OS to compute the quantiles) and Nc1 : : : Ncn are the number of unique
levels for the active classification variables of the given type.
Clearly, the memory requirements of the levels overwhelm the levels of the classification variables. For
information about how to adjust your computation resource parameters, see the SAS documentation for your
operating environment.
Another way to enhance performance is by carefully applying the TYPES or WAYS statement, limiting the
computations to only those combinations of classification variables that you are interested in.
Statistical Computations
Computation of Moment Statistics
PROC HPSUMMARY uses single-pass algorithms to compute the moment statistics (such as mean, variance,
skewness, and kurtosis). See the section Keywords and Formulas on page 153 for the statistical formulas.
The computational details for confidence limits, hypothesis test statistics, and quantile statistics follow.
Confidence Limits
With the statistic-keywords CLM, LCLM, and UCLM, you can compute confidence limits for the mean. A
confidence limit is a range (constructed around the value of a sample statistic) that contains the corresponding
true population value with given probability (ALPHA=) in repeated sampling. A two-sided 100.1 /%
confidence interval for the mean has upper and lower limits
xN t.1
s
n
=2I n 1/ p
q
P
N 2 and t.1
where s D n 1 1 .xi x/
with n 1 degrees of freedom.
A one-sided 100.1
s
n
s
1/ p
n
I n 1/ p
(upper)
xN
I n
(lower)
A two-sided 100.1
s
n 1
s
2
.1 =2I n
is the .1
=2/ critical value of the Students t statistic
/% confidence interval is computed as
xN C t.1
t.1
=2I n 1/
/% confidence interval for the standard deviation has lower and upper limits
s
n 1
; s
2
.=2I n 1/
1/
where 2.1 =2I n 1/ and 2.=2I n 1/ are the .1 =2/ and =2 critical values of the chi-square statistic with
n 1 degrees of freedom. A one-sided 100.1 /% confidence interval is computed by replacing =2 with
.
A 100.1 /% confidence interval for the variance has upper and lower limits that are equal to the squares
of the corresponding upper and lower limits for the standard deviation.
If you use the WEIGHT statement or the WEIGHT= option in a VAR statement and the default value of the
VARDEF= option (which is DF), the 100.1 /% confidence interval for the weighted mean has upper and
lower limits
yNw t.1
sw
=2/ v
u
n
uX
t
wi
i D1
where yNw is the weighted mean, sw is the weighted standard deviation, wi is the weight for the ith observation,
and t.1 =2/ is the .1 =2/ critical value for the Students t distribution with n 1 degrees of freedom.
Students t Test
PROC HPSUMMARY calculates the t statistic as
tD
xN 0
p
s= n
where xN is the sample mean, n is the number of nonmissing values for a variable, and s is the sample standard
deviation. Under the null hypothesis, the population mean equals 0 . When the data values are approximately
normally distributed, the probability under the null hypothesis of a t statistic as extreme as, or more extreme
than, the observed value (the p-value) is obtained from the t distribution with n 1 degrees of freedom. For
large n, the t statistic is asymptotically equivalent to a z test.
When you use the WEIGHT statement or the WEIGHT= option in a VAR statement and the default value of
the VARDEF= option (which is DF), the Students t statistic is calculated as
tw D
yNw 0
,v
uX
u n
sw t
wi
i D1
where yNw is the weighted mean, sw is the weighted standard deviation, and wi is the weight for the ith
observation. The tw statistic is treated as having a Students t distribution with n 1 degrees of freedom.
If you specify the EXCLNPWGT option in the PROC HPSUMMARY statement, then n is the number of
nonmissing observations when the value of the WEIGHT variable is positive. By default, n is the number of
nonmissing observations for the WEIGHT variable.
Quantiles
The options QMETHOD=, QNTLDEF=, and QMARKERS= determine how PROC HPSUMMARY calculates quantiles. The QNTLDEF= option deals with the mathematical definition of a quantile. See the section
Quantile and Related Statistics on page 157. The QMETHOD= option specifies how PROC HPSUMMARY
handles the input data: When QMETHOD=OS, PROC HPSUMMARY reads all data into memory and sorts
it by unique value. When QMETHOD=P2, PROC HPSUMMARY accumulates all data into a fixed sample
size that is used to approximate the quantile.
If data set A has 100 unique values for a numeric variable X and data set B has 1,000 unique values for
numeric variable X, then QMETHOD=OS for data set B requires 10 times as much memory as it does for data
set A. If QMETHOD=P2, then both data sets A and B require the same memory space to generate quantiles.
Results F 151
The QMETHOD=P2 technique is based on the piecewise-parabolic (P 2 ) algorithm invented by Jain and
Chlamtac (1985). P 2 is a one-pass algorithm to determine quantiles for a large data set. It requires a fixed
amount of memory for each variable for each level within the type. However, using simulation studies,
reliable estimations of some quantiles (P1, P5, P95, P99) cannot be possible for some data sets such as data
sets with heavily tailed or skewed distributions.
If the number of observations is less than the QMARKERS= value, then QMETHOD=P2 produces the
same results as QMETHOD=OS when QNTLDEF=5. To compute weighted quantiles, you must use
QMETHOD=OS.
Results
Missing Values
PROC HPSUMMARY excludes missing values for the analysis variables before calculating statistics. Each
analysis variable is treated individually; a missing value for an observation in one variable does not affect the
calculations for other variables. PROC HPSUMMARY handles missing values as follows:
If a classification variable has a missing value for an observation, then PROC HPSUMMARY excludes
that observation from the analysis unless you use the MISSING option in the PROC statement or
CLASS statement.
If a FREQ variable value is missing or nonpositive, then PROC HPSUMMARY excludes the observation from the analysis.
If a WEIGHT variable value is missing, then PROC HPSUMMARY excludes the observation from the
analysis.
PROC HPSUMMARY tabulates the number of the missing values. Before the number of missing values
are tabulated, PROC HPSUMMARY excludes observations with frequencies that are nonpositive when you
use the FREQ statement and observations with weights that are missing or nonpositive (when you use the
EXCLNPWGT option) when you use the WEIGHT statement. To report this information in the procedure
output use the NMISS statistic-keyword in the PROC HPSUMMARY statement.
The N Obs Statistic

By default when you use a CLASS statement, PROC HPSUMMARY displays an additional statistic called
N Obs. This statistic reports the total number of observations or the sum of the observations of the FREQ
variable that PROC HPSUMMARY processes for each class level. PROC HPSUMMARY might omit
observations from this total because of missing values in one or more classification variables. Because of this
action and the exclusion of observations when the weight-variable (specified in the WEIGHT statement or in
the WEIGHT= option in the VAR statement) contains missing values, there is not always a direct relationship
between N Obs, N, and NMISS.
In the output data set, the value of N Obs is stored in the _FREQ_ variable.
Output Data Set

PROC HPSUMMARY creates one output data set. The procedure does not print the output data set. Use the
PRINT procedure, the REPORT procedure, or another SAS reporting tool to display the output data set.
N OTE : By default, the statistics in the output data set automatically inherit the analysis variables format
and label. However, statistics computed for N, NMISS, SUMWGT, USS, CSS, VAR, CV, T, PROBT, PRT,
SKEWNESS, and KURTOSIS do not inherit the analysis variables format because this format can be invalid
for these statistics.
The output data set can contain these variables:
the variables specified in the CLASS statement.
the variable _TYPE_ that contains information about the classification variables. By default _TYPE_
is a numeric variable. If you specify CHARTYPE in the PROC statement, then _TYPE_ is a character
variable. When you use more than 32 classification variables, _TYPE_ is automatically a character
variable.
the variable _FREQ_ that contains the number of observations that a given output level represents.
the variables requested in the OUTPUT statement that contain the output statistics and extreme values.
the variable _STAT_ that contains the names of the default statistics if you omit statistic keywords.
The value of _TYPE_ indicates which combination of the classification variables PROC HPSUMMARY
uses to compute the statistics. The character value of _TYPE_ is a series of zeros and ones, where each value
of one indicates an active classification variable in the type. For example, with three classification variables,
PROC HPSUMMARY represents type 1 as 001, type 5 as 101, and so on.
Usually, the output data set contains one observation per level per type. However, if you omit statistickeywords in the OUTPUT statement, then the output data set contains five observations per level (six if you
specify a WEIGHT variable). Therefore, the total number of observations in the output data set is equal to
the sum of the levels for all the types that you request multiplied by 1, 5, or 6, whichever is applicable.
If you omit the CLASS statement (_TYPE_= 0), then there is always exactly one level of output per output
data set. If you use a CLASS statement, then the number of levels for each type that you request has an upper
bound equal to the number of observations in the input data set. By default, PROC HPSUMMARY generates
all possible types. In this case the total number of levels for each output data set has an upper bound equal to
m .2k
1/ n C 1
where k is the number of classification variables, n is the number of observations in the input data set, and m
is 1, 5, or 6.
PROC HPSUMMARY determines the actual number of levels for a given type from the number of unique
combinations of each active classification variable. A single level consists of all input observations whose
formatted class values match.
Table 10.8 shows the values of _TYPE_ and the number of observations in the data set when you specify one,
two, and three classification variables.
Keywords and Formulas F 153
Table 10.8 The Effect of Classification Variables on the

OUTPUT Data Set
CLASS Variables
C
0
0
0
0
1
1
1
1
0
0
1
1
0
0
1
1
_WAY_
_TYPE_
0
1
1
2
1
2
2
3
0
1
2
3
4
5
6
7
A
0
1
0
1
0
1
0
1
Character binary
equivalent of
_TYPE_
(CHARTYPE
option in the
PROC
HPSUMMARY
statement)
Subgroup
defined by
of this _TYPE_ and
_WAY_ in the data set
Total
A
B
A*B
C
A*C
B*C
A*B*C
1
a
b
a*b
c
a*c
b*c
a*b*c
Total number of
observations in
the data set
1+a
1+a+b+a*b
1+a+b+a*b+c
+a*c+b*c+a*b*c
a,b,c = number of levels of A, B, C,

respectively
Keywords and Formulas

Simple Statistics
The HPSUMMARY procedure uses a standardized set of keywords to refer to statistics. You specify these
keywords in SAS statements to request the statistics to be displayed or stored in an output data set.
In the following notation, summation is over observations that contain nonmissing values of the analyzed
variable and, except where shown, over nonmissing weights and frequencies of one or more:
xi
is the nonmissing value of the analyzed variable for observation i.
fi
is the frequency that is associated with xi if you use a FREQ statement. If you omit the FREQ statement,
then fi D 1 for all i.
wi
is the weight that is associated with xi if you use a WEIGHT statement. The HPSUMMARY procedure
automatically excludes the values of xi with missing weights from the analysis.
By default, the HPSUMMARY procedure treats a negative weight as if it is equal to 0. However, if
you use the EXCLNPWGT option in the PROC HPSUMMARY statement, then the procedure also
excludes those values of with nonpositive weights.
If you omit the WEIGHT statement, then wi D 1 for all i.
n
P
is the number of nonmissing values of xi , fi . If you use the EXCLNPWGT option and the WEIGHT
statement, then n is the number of nonmissing values with positive weights.
xN
is the mean
.X
X
wi xi
wi
s2
is the variance
1X
wi .xi
d
x/
N 2
where d is the variance divisor (the VARDEF= option) that you specify in the PROC HPSUMMARY
statement. Valid values are as follows:
When VARDEF=
N
DF
WEIGHT | WGT
WDF
d equals
n
n
P 1
Pi wi
i wi
The default is DF.

zi
is the standardized variable
.xi
x/=s
N
PROC HPSUMMARY calculates the following simple statistics:

number of missing values
number of nonmissing values
number of observations
sum of weights
mean
sum
extreme values
minimum
maximum
range
uncorrected sum of squares

corrected sum of squares
variance
standard deviation
standard error of the mean
coefficient of variation
skewness
kurtosis
confidence limits of the mean
median
mode
percentiles/deciles/quartiles
t test for mean=0
The standard keywords and formulas for each statistic follow. Some formulas use keywords to designate the
corresponding statistic.
Descriptive Statistics
The keywords for descriptive statistics are as follows:
CSS
is the sum of squares corrected for the mean, computed as
X
wi .xi x/
N 2
CV
is the percent coefficient of variation, computed as
.100s/=xN
KURTOSIS | KURT
is the kurtosis, which measures heaviness of tails. When VARDEF=DF, the kurtosis is computed as
c4n
zi4
where c4n is
3 .n 1/2
.n 2/ .n 3/
n.nC1/
.n 1/.n 2/.n 3/
When VARDEF=N, the kurtosis is computed as

1X 4
zi 3
n
The formula is invariant under the transformation wi D zwi , z > 0. When you use VARDEF=WDF
or VARDEF=WEIGHT, the kurtosis is set to missing.
MAX
is the maximum value of xi .
MEAN
is the arithmetic mean x.
N
MIN
is the minimum value of xi .
MODE
is the most frequent value of xi .
N OTE : When QMETHOD=P2, PROC HPSUMMARY does not compute MODE.
N
is the number of xi values that are not missing. Observations with fi < 1 and wi equal to missing or
wi 0 (when you use the EXCLNPWGT option) are excluded from the analysis and are not included
in the calculation of N.
NMISS
is the number of xi values that are missing. Observations with fi < 1 and wi equal to missing or
wi 0 (when you use the EXCLNPWGT option) are excluded from the analysis and are not included
in the calculation of NMISS.
NOBS
is the total number of observations and is calculated as the sum of N and NMISS. However, if you
use the WEIGHT statement, then NOBS is calculated as the sum of N, NMISS, and the number of
observations excluded because of missing or nonpositive weights.
RANGE
is the range and is calculated as the difference between maximum value and minimum value.
SKEWNESS | SKEW
is skewness, which measures the tendency of the deviations to be larger in one direction than in the
other. When VARDEF=DF, the skewness is computed as
X
c3n
zi3
where c3n is
n
.
.n 1/.n 2/
When VARDEF=N, the skewness is computed as

1X 3
zi
n
The formula is invariant under the transformation wi D zwi , z > 0. When you use VARDEF=WDF
or VARDEF=WEIGHT, the skewness is set to missing.
STDDEV | STD
is the standard deviation s and is computed as the square root of the variance, s 2 .
STDERR | STDMEAN
is the standard error of the mean, computed as
s
pP
wi
when VARDEF=DF, which is the default. Otherwise, STDERR is set to missing.
SUM
is the sum, computed as
X
wi xi
SUMWGT
is the sum of the weights, W, computed as
X
wi
USS
is the uncorrected sum of squares, computed as
X
wi xi2
VAR
is the variance s 2 .
Quantile and Related Statistics

The keywords for quantiles and related statistics are as follows:
MEDIAN
is the middle value.
Pn
is the nth percentile. For example, P1 is the first percentile, P5 is the fifth percentile, P50 is the 50th
percentile, and P99 is the 99th percentile.
Q1
is the lower quartile (25th percentile).
Q3
is the upper quartile (75th percentile).
QRANGE
is interquartile range and is calculated as
Q3
Q1
You use the QNTLDEF= option to specify the method that the HPSUMMARY procedure uses to compute
percentiles. Let n be the number of nonmissing values for a variable, and let x1 ; x2 ; : : : ; xn represent the
ordered values of the variable such that x1 is the smallest value, x2 is the next smallest value, and xn is the
largest value. For the tth percentile between 0 and 1, let p D t =100. Then define j as the integer part of np
and g as the fractional part of np or .n C 1/p, so that
np D j C g
.n C 1/p D j C g
when QNTLDEF=1, 2, 3, or 5
when QNTLDEF=4
Here, QNTLDEF= specifies the method that the procedure uses to compute the tth percentile, as shown in
Table 10.10.
When you use the WEIGHT statement, the tth percentile is computed as
8
i
X
/
.x
if
wj D pW
C
x
i C1
<2 i
j D1
yD
i
i C1
X
X
x
if
wj < pW <
wj
: i C1
j D1
j D1
where wj is the weight associated with xi and W D
n
X
wi is the sum of the weights. When the obser-
i D1
vations have identical weights, the weighted percentiles are the same as the unweighted percentiles with
QNTLDEF=5.
Table 10.10 Methods for Computing Quantile Statistics
QNTLDEF=
1
Description
Weighted average at xnp
Observation numbered closest to np
Empirical distribution function
Weighted average aimed at x.nC1/p
Empirical distribution function with

averaging
Formula
y D .1 g/xj Cgxj C1
where x0 is taken to be x1
y D xi
if g 21
y D xj
if g D 12 and j is even
y D xj C1
if g D 21 and j is odd
where i is the integer part of np C 12
y D xj
if g D 0
y D xj C1
if g > 0
y D .1 g/xj Cgxj C1
where xnC1 is taken to be xn
y D 21 .xj C xj C1 /
if g D 0
y D xj C1
if g > 0
Hypothesis Testing Statistics

The keywords for hypothesis testing statistics are as follows:
T
is the Students t statistic to test the null hypothesis that the population mean is equal to 0 and is
calculated as
xN 0
.pP
s
wi
By default, 0 is equal to zero. You must use VARDEF=DF, which is the default variance divisor;
otherwise T is set to missing.
By default, when you use a WEIGHT statement, the procedure counts the xi values with nonpositive
weights in the degrees of freedom. Use the EXCLNPWGT option in the PROC HPSUMMARY
statement to exclude values with nonpositive weights.
PROBT | PRT
is the two-tailed p-value for the Students t statistic, T, with n 1 degrees of freedom. This value is the
probability under the null hypothesis of obtaining a more extreme value of T than is observed in this
sample.
Confidence Limits for the Mean

The keywords for confidence limits are as follows:
CLM
is the two-sided confidence limit for the mean. A two-sided 100.1
the mean has upper and lower limits
xN t.1
/ percent confidence interval for
s
=2I n 1/ pP
wi
P
where s is n 1 1 .xi x/
N 2 , t.1 =2I n 1/ is the .1 =2/ critical value of the Students t statistic
with n 1 degrees of freedom, and is the value of the ALPHA= option which by default is 0.05.
Unless you use VARDEF=DF (which is the default variance divisor), CLM is set to missing.
LCLM
is the one-sided confidence limit below the mean. The one-sided 100.1 / percent confidence interval
for the mean has the lower limit
xN
t.1
s
I n 1/ pP
wi
Unless you use VARDEF=DF (which is the default variance divisor), LCLM is set to missing.
UCLM
is the one-sided confidence limit above the mean. The one-sided 100.1
for the mean has the upper limit
xN C t.1
/ percent confidence interval
s
I n 1/ pP
wi
Unless you use VARDEF=DF (which is the default variance divisor), UCLM is set to missing.
Data Requirements for the HPSUMMARY Procedure

The following are the minimal data requirements to compute unweighted statistics and do not describe
recommended sample sizes. Statistics are reported as missing if VARDEF=DF (the default) and the following
requirements are not met:
N and NMISS are computed regardless of the number of missing or nonmissing observations.
SUM, MEAN, MAX, MIN, RANGE, USS, and CSS require at least one nonmissing observation.
VAR, STD, STDERR, CV, T, PRT, and PROBT require at least two nonmissing observations.
SKEWNESS requires at least three nonmissing observations.
KURTOSIS requires at least four nonmissing observations.
SKEWNESS, KURTOSIS, T, PROBT, and PRT require that STD is greater than zero.
CV requires that MEAN is not equal to zero.
CLM, LCLM, UCLM, STDERR, T, PRT, and PROBT require that VARDEF=DF.
References
Jain, R. and Chlamtac, I. (1985), The P 2 Algorithm for Dynamic Calculation of Quantiles and Histograms
without Storing Observations, Communications of the ACM, 28, 10761085.
Subject Index
binning information table, 55
class level
HPSAMPLE procedure, 124
CODE statement, 49
computation, 51
corrected sums of squares and crossproducts, 71
correlation coefficients, 67
limited combinations of, 74
printing, for each variable, 71
suppressing probabilities, 71
covariances, 71
displayed output
HPIMPUTE procedure, 116
formula, 51
FREQ statement, 49
frequency information table
frequency variable
HPBIN procedure
WOE example, 65
HPCORR procedure
concepts, 74
details, 74
examples, 78
multithreading, 73, 76
ODS table names, 78
output, 76
output data sets, 77
overview, 67
syntax, 70
task tables, 71
HPDMDB procedure, 83
multithreading, 90
HPDS2 procedure, 95
input data sets, 100
multithreading, 101
displayed output, 116
multithreading, 114
ODS table names, 116
HPREG procedure
introductory example, 120
class level, 124
displayed output, 124
frequency information table, 124
ODS table names, 125
performance information, 124
single threading, 123
timing table, 124
HPSUMMARY procedure, 131
analysis variable, 145
classification variables, 140
confidence level, 136
missing values, 137
multithreading, 143
ID, 49
ID statement, 49
information value (IV) table, 55
INPUT statement, 49
PROC HPBIN variables, 49
mapping ODS table, 56
measures of association, 78
nonparametric, 67
multithreading
HPCORR procedure, 73, 76
HPDS2 procedure, 101
NObs ODS table, 56
nonparametric measures of association, 67
ODS (Output Delivery System)
HPCORR procedure and, 78
ODS tables, 55
options summary
PROC HPSUMMARY statement, 136
output
output table details, 54
Pearson correlation statistics, 67
example, 78, 80
in output data set, 71
Pearson product-moment correlation, 71, 74, 78
162 F Subject Index
Pearson weighted product-moment correlation, 74

probability values, 75
suppressing, 71
performance information
performance information ODS table, 56
PERFORMANCE statement, 50
PROC HPBIN, 41
PROC HPBIN binning methods
bucket, 42
pseudoquantile, 43
quantile, 42
Winsorized, 42
PROC HPBIN features, 43
PROC HPBIN statement, 47
BINS_META= data set, 47
COMPUTEQUANTILE option, 48
COMPUTESTATS option, 48
input data, 48
method option, 47
NOPRINT option, 48
NUMBIN= option, 48
OUTPUT= option, 48
WOE option, 48, 49
PROC HPBIN syntax, 46
procedure output, 54
quantiles and extremes ODS table, 56
single threading
singularity of variables, 71
standard deviation, 71
summary statistics ODS table, 56
sums of squares and crossproducts, 71
timing table
trimmed statistics ODS table, 56
variances, 71
weight of evidence (WOE) ODS table, 56
Winsor statistics ODS table, 56
Syntax Index
ALLTYPES option
ALLWAYS option
ALPHA option
PROC HPCORR statement, 71
ALPHA= option
analysis variable
VAR statement (HPSUMMARY), 145
AUTONAME option
OUTPUT statement (HPSUMMARY), 143
BEST= option
CHARTYPE= option
CLASS statement, see HPDMDB procedure, see
HPSAMPLE procedure, see HPSUMMARY
procedure
classification variables
CLASS statement (HPSUMMARY), 140
CODE statement, 49
COMMIT= option
PERFORMANCE statement (high-performance
analytical procedures), 36
CORR procedure, WITH statement, 74
COV option
CSSCP option
DATA statement, see HPDS2 procedure
DATA= option
PROC HPDS2 statement, 100
PROC HPIMPUTE statement, 112
DETAILS option
ENDDATA statement, see HPDS2 procedure

EXCLNPWGT option
EXCLNPWGTS option, see EXCLNPWGT option
FMTLIBXML= option
FREQ statement, 49, see HPDMDB procedure, see
HPSUMMARY procedure
HPCORR procedure, 73
frequency variable
FREQ statement (HPSUMMARY), 141
GRIDHOST= option
GRIDMODE= option
GRIDTIMEOUT= option
GROUPINTERNAL option
CLASS statement (HPSUMMARY), 140
high-performance analytical procedures,
COMMIT= option, 36
DETAILS option, 37
GRIDHOST= option, 37
GRIDMODE= option, 37
GRIDTIMEOUT= option, 37
HOST= option, 37
INSTALL= option, 37
INSTALLLOC= option, 37
LASR= option, 38
LASRSERVER= option, 38
MODE= option, 37
NNODES= option, 38
NODES= option, 38
NTHREADS= option, 39
THREADS= option, 39
TIMEOUT= option, 37
HOST= option
164 F Syntax Index

HPBIN procedure
TARGET statement, 50
HPCORR procedure
syntax, 70
HPCORR procedure, FREQ statement, 73
HPCORR procedure, PERFORMANCE statement, 73
HPCORR procedure, PROC HPCORR statement, 71
ALPHA option, 71
BEST= option, 71
COV option, 71
CSSCP option, 71
DATA= option, 71
EXCLNPWGT option, 72
NOCORR option, 72
NOMISS option, 72
NOPRINT option, 72
NOPROB option, 72
NOSIMPLE option, 72
OUT= option, 72
OUTP= option, 72
RANK option, 72
SSCP option, 72
VARDEF= option, 72
HPCORR procedure, VAR statement, 73
HPCORR procedure, WEIGHT statement, 74
CLASS statement, 89
FREQ statement, 90
PROC HPDMDB statement, 88
syntax, 88
VAR statement, 90
WEIGHT statement, 91
HPDMDB procedure, CLASS statement, 89
HPDMDB procedure, FREQ statement, 90
HPDMDB procedure, PERFORMANCE statement, 90
HPDMDB procedure, PROC HPDMDB statement, 88
HPDMDB procedure, VAR statement, 90
HPDMDB procedure, WEIGHT statement, 91
DATA statement, 100
ENDDATA statement, 101
QUIT statement, 101
RUN CANCEL statement, 102
RUN statement, 102
syntax, 100
HPDS2 procedure, DATA statement, 100
HPDS2 procedure, ENDDATA statement, 101
HPDS2 procedure, PERFORMANCE statement, 101
HPDS2 procedure, PROC HPDS2 statement

DATA= option, 100
FMTLIBXML= option, 100
IN= option, 100
OUT= option, 100
OUTPUT= option, 100
HPDS2 procedure, QUIT statement, 101
HPDS2 procedure, RUN CANCEL statement, 102
HPDS2 procedure, RUN statement, 102
HPDS2C procedure, PROC HPDS2 statement, 100
CODE statement, 113
FREQ statement, 113
ID statement, 113
IMPUTE statement, 113
INPUT statement, 114
syntax, 112
HPIMPUTE procedure, PROC HPIMPUTE statement
DATA= option, 112
OUT= option, 112
CLASS statement, 123
PROC HPSAMPLE statement, 121
syntax, 121
TARGET statement, 123
VAR statement, 123
HPSAMPLE procedure, CLASS statement, 123
HPSAMPLE procedure, PERFORMANCE statement,
123
HPSAMPLE procedure, PROC HPSAMPLE
statement, 121
HPSAMPLE procedure, TARGET statement, 123
HPSAMPLE procedure, VAR statement, 123
CLASS statement, 140
FREQ statement, 141
OUTPUT statement, 141
syntax, 135
TYPES statement, 143
VAR statement, 144
WAYS statement, 145
WEIGHT statement, 146
HPSUMMARY procedure, CLASS statement, 140
GROUPINTERNAL option, 140
HPSUMMARY procedure, FREQ statement, 141
frequency variable, 141
HPSUMMARY procedure, OUTPUT statement, 141
AUTONAME option, 143
OUT= option, 141
Syntax Index F 165
output statistic specification, 141

HPSUMMARY procedure, PERFORMANCE
statement, 143
HPSUMMARY procedure, PROC HPSUMMARY
statement, 136
ALLTYPES option, 136
ALLWAYS option, 136
ALPHA= option, 136
CHARTYPE= option, 137
classification variables, 140
DATA= option, 137
EXCLNPWGT option, 137
MISSING option, 137
PCTLDEF= option, 138
QMARKERS= option, 137
QMETHOD= option, 137
QNTLDEF= option, 138
statistics keywords, 139
VARDEF= option, 138
HPSUMMARY procedure, TYPES statement, 143
requests, 144
HPSUMMARY procedure, VAR statement, 144
analysis variable, 145
WEIGHT= option, 145
HPSUMMARY procedure, WAYS statement, 145
list, 146
HPSUMMARY procedure, WEIGHT statement, 146
ID statement, 49
IMPUTE statement
IN= option
INPUT statement, 49
INSTALL= option
INSTALLLOC= option
LASR= option
LASRSERVER= option
list
WAYS statement (HPSUMMARY), 146
MISSING option
MODE= option

NNODES= option
NOCORR option
NODES= option
NOMISS option
NOPRINT option
NOPROB option
NOSIMPLE option
NTHREADS= option
OUT= option
OUTP= option
OUTPUT statement, see HPSUMMARY procedure
output statistic specification
OUTPUT= option
PCTLDEF= option
high-performance analytical procedures, 36
PROC HPBIN statement, 47
BINS_META= data set, 47
COMPUTEQUANTILE option, 48
COMPUTESTATS option, 48
input data, 48
method option, 47
NOPRINT option, 48
NUMBIN= option, 48
166 F Syntax Index
OUTPUT= option, 48
WOE option, 48, 49
PROC HPCORR statement, 71, see HPCORR
procedure
PROC HPDMDB statement, see HPDMDB procedure
PROC HPDS2 statement, see HPDS2 procedure
PROC HPIMPUTE statement
PROC HPSAMPLE statement, see HPSAMPLE
procedure
PROC HPSUMMARY statement, see HPSUMMARY
procedure
QMARKERS= option
QMETHOD= option
QNTLDEF= option
QUIT statement, see HPDS2 procedure
RANK option
requests
TYPES statement (HPSUMMARY), 144
RUN CANCEL statement, see HPDS2 procedure
RUN statement, see HPDS2 procedure
SSCP option
statistics keywords
syntax
TARGET statement, see HPSAMPLE procedure
HPBIN procedure, 50
THREADS= option
TIMEOUT= option

TYPES statement, see HPSUMMARY procedure
VAR statement, see HPDMDB procedure, see
HPSAMPLE procedure, see HPSUMMARY
procedure
VARDEF= option
WAYS statement, see HPSUMMARY procedure
WEIGHT statement, see HPDMDB procedure, see
HPSUMMARY procedure
WEIGHT= option
VAR statement (HPSUMMARY), 145
WITH statement
CORR procedure, 74

Base SAS 9.4 Procedures Guide High-Performance Procedures, Third Edition

Uploaded by

Copyright:

Available Formats

Base SAS 9.4 Procedures Guide High-Performance Procedures, Third Edition

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Base SAS 9.4 Procedures Guide High-Performance Procedures, Third Edition

Uploaded by

Copyright:

Available Formats

What is the purpose of this manual?

What is the purpose of this manual?

What procedures are covered in this manual?

What procedures are covered in this manual?

Base SAS 9.

Gain Greater Insight into Your

for additional books and resources.

Whats New in Base SAS High-Performance Procedures 13.2

Credits and Acknowledgments

Anne Baxter, Ed Huddleston

Tim Arnold, Melanie Gratton

Tao Wang, Taiping He

Whats New in Base SAS High-Performance

GRIDDATASERVER= Option in the PERFORMANCE Statement No Longer Needed

GRIDMODE= Option in the PERFORMANCE Statement No Longer Needed . . . .

Multiple Data Access Modes Supported in a Single Procedure Step . . . . . . . . . .

Running Asymmetrically Alongside SAP HANA . . . . . . . . . . . . . . . . . . . .

GRIDMODE= Option in the PERFORMANCE Statement No Longer Needed

4 F Chapter 1: Whats New in Base SAS High-Performance Procedures 13.2

Multiple Data Access Modes Supported in a Single Procedure Step

Running Asymmetrically Alongside SAP HANA

About This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Options Used in Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

SAS Technical Support Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Overview of Base SAS High-Performance Procedures

SAS High-Performance Analytics Product

SAS High-Performance Statistics

About This Book

is the standard type style used for most text.

is used to for matrices and vectors.

Options Used in Examples F 7

Options Used in Examples

SAS Technical Support Services

Shared Concepts and Topics

Determining Single-Machine Mode or Distributed Mode . . . . . . . . . . . . . . . .

Data Access Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Single-Machine Data Access Mode . . . . . . . . . . . . . . . . . . . . . . . . . . .

Distributed Data Access Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Determining the Data Access Mode . . . . . . . . . . . . . . . . . . . . . . . . . . .

Alongside-LASR Distributed Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Running High-Performance Analytical Procedures Alongside a SAS LASR Analytic Server

Starting a SAS LASR Analytic Server Instance . . . . . . . . . . . . . . . . . . . . .

Terminating a SAS LASR Analytic Server Instance . . . . . . . . . . . . . . . . . . .

Alongside-LASR Distributed Execution on a Subset of the Appliance Nodes . . . . . . . . .

Running High-Performance Analytical Procedures in Asymmetric Mode . . . . . . . . . . .

Running in Symmetric Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Running in Asymmetric Mode on One Appliance . . . . . . . . . . . . . . . . . . . .

Running in Asymmetric Mode on Distinct Appliances . . . . . . . . . . . . . . . . .

Alongside-HDFS Execution by Using the SASHDAT Engine . . . . . . . . . . . . .

Alongside-HDFS Execution by Using the Hadoop Engine . . . . . . . . . . . . . . .

Output Data Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Working with Formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10 F Chapter 3: Shared Concepts and Topics

Controlling the Execution Mode with Environment Variables and

A specification in the PERFORMANCE statement overrides a specification of an environment variable

12 F Chapter 3: Shared Concepts and Topics

Determining Single-Machine Mode or Distributed Mode

proc hplogistic data=simData;

Figure 3.1 shows the results from the analysis.