FROM EXCEL TO

MACHINE LEARNING
BOOTCAMP

YOUR TEAM'S EXCEL SKILLS WILL

UNLOCK ADVANCED ANALYTICS
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Introduction

IS EASY
I am a hands-on analytics professional and experienced leader. I mention this only to provide
credibility for what I am about to say next.

Over the years I have performed analytics on behalf of teams in Marketing, Product
Management, Finance, Customer Service, and Supply Chain.

In the majority of cases, these analyses could be performed by the business subject matter
experts (SMES). The techniques I use are not that hard to learn. This includes business SMEs
acquiring machine learning skills to analyze their data.

I know. It sounds a bit crazy.

LINEAR
My philosophy is to let my content do the talking - which is why this document exists.

This document will demonstrate the following:

REGRESSION
1. Any business team with Excel skills can learn to analyze data with R programming.
2. Any business team with R skills can acquire valuable machine learning skills.

Analyzing data with machine learning is applicable to all aspects of business:

[Human Resources] - What factors are highly predictive of bad attrition?

[Product Management] - What feature usage(s) are highly predictive of a sticky SaaS customer?
[Customer Service] - Are there factors related to churn that customer service can address?
[Marketing] - What are the customer journey behaviors that predict conversion?
[Supply Chain] - What are the factors/behaviors that predict customer returns?

If your team wants to have more business impact using data, I invite you to keep reading.

-Dave

2
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Document Goals

IS EASY
This document is structured to accomplish the following goals:

1. To demonstrate that any business team with Excel skills can learn R programming.
2. To demonstrate that any business team can acquire machine learning skills.
3. To provide an outline of the 3-day bootcamp curriculum

Per the above, the document is structured as follows:

Part 1 illustrates how Excel knowledge maps directly to R programming.

Part 2 illustrates how approachable machine learning is to any business team.
Part 3 provides the outline of topics for the 3-day bootcamp.

LINEAR
A secondary benefit of this document is that you will acquire some actual machine learning
knowledge as the result of your reading.

REGRESSION

3
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Table of Contents

Part1 - Why Programming Is Easy

Excel Users Write Code................................7

Excel Code
R Code
Excel Skills Translate Directly to R

Excel and R Are Like Icebergs.....................10

It's All About the Tables...............................11

Tables Are Key
Tables in Excel and R
Cells of Data in Excel and R
Columns of Data in Excel and R

Throw in Some Functions...........................18

Using Functions in Excel and R
Common Functions

Awesome Data Visualizations...................21

Excel Data Visualizations
R Data Visualizations

4
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Table of Contents

Part 2 - Machine Learning for Any Professional

Classification Tree Intuition...................23

Trees Are Rules
Let's Build a Tree!

Overfitting Intuition................................31
The Bugbear of Machine Learning
The Model Is Good - Or Is It?
What Happened?
Model Tuning Intuition

Gini Impurity.............................................36
Impurity Intuition
Gini Impurity
Gini Impurity Example

5
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Table of Contents

Part 3 - Bootcamp Topic Outline

Day 1 - R Programming Made Easy...................................................39

Days 2 & 3 - Introduction to Machine Learning With R...............42

Wrap Up

Ready to Empower Your Team?.......................................................47

About the Author.................................................................................48

6
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Excel Users Write Code

Excel Code

While most Excel users don't think of it this way, they spend a lot of time writing and
debugging code in Excel. In fact, Microsoft Excel is by far and away the world's most
popular programming environment.

Take the image below as an example. The user is using Excel's AVERAGE function to
calculate the average of a column (i.e., Petal.Width) of a table (i.e., iris_data) in a
worksheet.

LINEAR
Once the user hits the <enter> key, Excel attempts to interpret the instructions in the
cell and perform the desired operation. If Excel doesn't understand what the user
REGRESSION
typed, it reports an error.

That's coding!

7
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

R Code

While it isn't the only way to code in Excel, calling Excel functions as depicted on the
previous page is by far the most common. When using Excel in this way, Excel is
operating as a code interpreter.

Using R as an interpreter is very common. The R user types some code and hits the
<enter> key. R then tries to interpret the code, throwing an error if doesn't
understand what was typed by the user.

In this way Excel and R are very similar, but it doesn't stop there. Even the code is
very similar!
LINEAR
The image below depicts the same scenario as on the previous page, but using R
instead. REGRESSION
As depicted, the user is calculating the average (mean is just another name
for the average) of the Petal.Width column of the iris_data table.

8
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Excel Skills Directly Translate to R

While this example might seem simple, it demonstrates why R is the fastest, easiest
way for ANY team to unlock advanced analytics.

As you will see through the rest of this document, Excel is a powerful analytical tool
with many concepts and skills that need to be mastered to use Excel effectively.

This knowledge is directly applicable to R.

This knowledge makes the learning process an exercise in mapping Excel skills to R.

LINEAR
REGRESSION

9
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Excel and R Are Like Icebergs

While cliche, the iceberg metaphor is a powerful way to conceptualize the extent
Excel knowledge maps to R.

As depicted below, Excel features above the water line (e.g., Pivot Tables) only
scratch the surface of Excel's capabilities. However, these feature represent the bulk
of Excel's use in practice.

Another similarity between Excel and R is the "choose your own adventure" aspect
of the technologies. Just as many Excel users never learn Power Query, not every R
user needs to learn statistical analysis to be effective in their work.
LINEAR
REGRESSION
Tables Data Frames
Common Functions Common Functions
Pivot Tables dplyr
Charts ggplot2

Power Query dplyr

Analysis ToolPak Linear Regression
Solver Logistic Regression

Statistical Functions Statistical Functions

DAX Market Basket Analysis
VBA Machine Learning

10
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

It's All About the Tables

Tables Are Key
IS EASY
Using Microsoft Excel is all about working with tables of data. You filter tables, you
sort tables, you pivot tables, etc. You can think of tables as one of the fundamental
objects in Excel that you create and manipulate to conduct analyses.

Excel tables can also be thought of as container objects. Tables contain rows, columns,
cells, data formats, etc.

You probably can see where I'm going with this already.

LINEAR
When analyzing data with R, it's all about the tables - just like Excel.

REGRESSION
Once again, your Excel knowledge directly translates to R.

11
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Tables in Excel and R

The image below demonstrates how Excel tables are objects. For example, every
table in Excel has a name - whether you explicitly name a table or not. Table names
allow you to directly access/manipulate tables using Excel code.

12
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Things work in R exactly the same way. Tables of data in R (known as "data frames")
have names just like Excel tables so that you can write R code to access/manipulate
tables of data.

[Excel Code] =AVERAGE(iris.data[Petal.Width])

[R Code] mean(iris.data$Petal.Width)

13
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Cells of Data in Excel and R

Working with cells of data is very common in Excel. It is useful to think of cells as
objects contained within tables - as depicted below. Once again, you use Excel code
to access cells.

14
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Although the R code to access/manipulate cells of data is slightly different,

conceptually it matches what Excel users do all the time.

15
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Columns of Data in Excel and R

Excel code supports different ways of accessing columns of data within tables. Two
examples:

Providing a worksheet-based cell range

Providing object names

16
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Once again, Excel knowledge maps directly to R code as illustrated below.

Notice how similar the actual R code is to Excel when using object names.

17
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Throw in Some Functions

Using Functions in Excel and R

The bulk of code Excel users write call functions. Often, these function calls are nested
and can be difficult to debug (again, that's coding!).

A common Excel scenario is depicted below - creating a new column of data

(IsSetosa) by invoking a function on another column of data (Species) within the
same table (iris.data).

In this utterly contrived example, Excel's mighty IF function (a commonly nested

Excel function) is being used.

18
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

The contrived example from the previous page is repeated using R code.

A IsSetosa column is being added to the iris.data table (or data frame) and populated
with new data derived from the existing Species column.

First, notice how the workflow is exactly the same as in Excel - only everything is
done in code with R.

Second, notice how similar the R ifelse function call is to Excel code.

18
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Common Functions

Like Excel, R comes out of the box with many, many functions to work with columns
of data.

Many of the R functions share the same name with the corresponding Excel
function. In other cases, mapping your Excel knowledge to R is straightforward, as
depicted below.

20
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Awesome Data Visualizations

Excel Data Visualizations

Excel is a great data visualization tool, supporting many ways to analyze data
visually.

21
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

R Data Visualizations

R is renowned for it's ability to create print-quality data visualizations.

R also easily produces data visualizations that are difficult, or not possible, to do with
out of the box Excel features.

Analyzing data visually is one of R's specialities.

22
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Classification Tree Intuition

In machine learning there are many techniques for building predictive models. The
technical term for these techniques is the word algorithm. Think of an algorithm as a
kind of recipe that the machine (i.e., the computer) follows to build the predictive
model.

Turns out that one of the most useful machine learning algorithms for business data
is also one of the simplest to learn - decision trees. Almost everyone is familiar with a
decision tree. You start at the top of the tree and move down each decision in the
tree until you reach the bottom.

A simple decision tree is depicted below.

23
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Trees Are Rules

Machine learning algorithms learn from data. In the case of decision trees, the
algorithm learns rules from the data.

Some sample data that was used to create the tree on the previous page is depicted
below. When we apply the decision tree algorithm to the data, the algorithm tries to
learn the rules to accurately predict the label using the features in the data set.

When a decision tree is asked to learn how to predict labels, it is known as a

classification tree.

24
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Most of us are familiar with decision trees being depicted in a graphical form as
shown on page 23. However, you can also represent a decision tree using text.

The tree learned from the data listed on the previous page is represented below as
text.

What is clear from this representation is that trees are collections of rules used to
make predictions.

The textual rules below clearly show what the algorithm learned from the data in
terms of predicting income levels (i.e., the label).

For example, using the rules below, a person that had an occupation of "Adm-
clerical" is predicted to have an annual income of less than or equal to $50,000 USD.

25
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Machine learning algorithms work by pursuing some sort of objective. Think of the
objective as how the algorithm determines if it has learned all that it can from the
data.

In the case of classification trees, the objective is to minimize impurity. Later you will
learn about the math of impurity, for now let's focus on the intuition.

A collection of data is pure when all the labels are the same. Reflexively, a collection
of data is impure when the labels are not all the same.

26
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Let's Build a Tree!

In this contrived example we have 10 observations (or rows) and 2 features of data.

The classification tree algorithm loves a single feature with lots of observations and
only a single label value.

This obsessive love is known as being greedy...

27
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

This is the first conceptual step of the classification tree algorithm learning from the
contrived data set.

Graphically, we can represent what the algorithm has learned so far as depicted
below.

However, not all the data has been used in the algorithm's learning process, so we
are not done yet!

28
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

As depicted below, the tree so far has only used 40% of the data for learning.

The observations used so far have been hidden by bars to clearly show what data is
left for the algorithm to use.

Of the data left for use by the algorithm, a pure split using the relationship feature is
highlighted.

The algorithm chooses this feature next.

29
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

The math of impurity tells the algorithm that after using the relationship feature
value of Wife, there are no valid ways of splitting the data further.

As such, the algorithm completes the tree as depicted below.

Viola! You now have a conceptual understanding of how classification trees learn
from data.

30
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Overfitting Intuition
The Bugbear of Machine Learning

As a business professional applying machine learning to your data, you must be

paranoid about overfitting.

Simply put, overfitting is where your model's predictions are much less "accurate"
when presented with new data.

The full bootcamp covers overfitting in depth, this document will focus on the
intuition.

Take the tree from the previous page as an example.

31
FROM EXCEL TO MACHINE LEARNING BOOTCAMP

The Model Is Good!

Considering the data used

to train the classification
tree model, things look
awesome!

When we look at the

original data and the tree
below, we see that the
model correctly predicts 9
out of 10 (i.e., 90%) of the
labels correctly!

32
FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Or Is It?

Take the data to the left that

was not used in training the
model. This is "new data."

When the model below is

used to make predictions for
this data it gets 9 out of 9 (i.e.,
100%) of the labels wrong!

This is the essence of

overfitting.

33
FROM EXCEL TO MACHINE LEARNING BOOTCAMP

What Happened?

Given the relatively small

amount of data to the left, the
tree is far too specialized.

For example, the model

assumes that any person with
a relationship of Wife makes
under $50,000 USD.

While that doesn't make sense

to us humans, it makes perfect
sense to the greedy algorithm
based on the data and how the
model was trained.

34
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Model Tuning Intuition

The bootcamp has robust coverage of model tuning. For the purposes of this
document, the focus will be on the intuition behind tuning machine learning
models.

Conceptually, think of model tuning like tuning your automobile for optimal
performance.

In terms of decision trees...

The decision tree "engine" has a number of settings that you can control as the
machine learning practitioner

These settings are called hyperparameters and are typically changed as needed for
optimal decision tree performance - just like a mechanic can change the engine
settings in your automobile for optimal performance.

In the case of decision trees, tuning manifests as controlling how specialized (i.e.,
how complex) a tree can become from the data used to train it.

In the case of the tree on the previous page, you would tune the hyperparameters
to make a less complex tree to pursue better predictions on new data.

35
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Gini Impurity
Impurity Intuition

The bootcamp has robust coverage of the mathematics of decision trees. Here's the
good news:

1. The mathematics are simple and accessible to any professional.

2. Thorough understanding mathematics is useful, but not required.

This document will focus on the intuition of impurity math, starting with a 2-label
(i.e., binary) case.

Using our example, we can think about "buckets" of labels and impurity as a
spectrum...

36
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Gini Impurity

Turns out there are several calculations that manifest the intuition of the previous
page.

The bootcamp uses the Gini impurity calculation.

Gini impurity is widely used and is the default calculation used by the R packages
taught in the bootcamp.

Don't panic!

While the math may appear intimidating, when you think about it in English, it is
really quite easy...

37
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Gini Impurity Example

As we know, the classification tree algorithm loves to split up the data into groups
where the labels are all the same (i.e., pure).

In other words, the classification tree algorithm's objective is to minimize impurity.

Using the math from the previous page we can now do some calculations and see
how pure buckets of labels have the smallest Gini impurity scores.

Notice that the worst case scenario is a 50/50 split of labels. This makes intuitive
sense when you have 2 labels - it is essentially a coin flip.

38
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Day 1 - R Programming Made Easy

Module 1 - Welcome
Bootcamp Expectations
The Bootcamp Data
Iris Data Set
Adult Census Income Data Set
Titanic Data Set

Module 2 - R and RStudio

The History of R
Similarities Between Excel & RStudio
Excel and R as Data Analysis Ecosystems
Excel Users Write Code
Excel Code Is Like R Code
Extending the Excel & R Ecosystems

Module 3 - Data Analysis Is All About Objects

Objects in Excel & R
It All Starts With Tables
Tables in Excel & R
Cells of Data
Columns of Data
Vectors of Data
Excel Cell Ranges
R Vectors
Data Types
Excel Data Formats
R Data Types
Math With Vectors
Excel-Style Vector Math
R Vector Math
39
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Module 4 - Filtering R Tables

Logical Filtering
Excel Logical Filtering
R Logical Filtering
Filtering with %in%
R Filtering
Filtering With Numbers & Names
Filtering With Direct Logic
Hands-on Lab #1

Module 5 - Basic Functions

Missing Data
Common Stats Functions
Mapping Excel Functions to R Functions
The R summary Function
Summary Stats in Excel
The data.frame Function
The cbind & rbind Functions
The aggregate Function
Excel Data Aggregation
R Data Aggregation
Hands-on Lab #2

Module 6 - Pivoting Data

The Mighty dplyr
Another Kind of Table - tibbles
Making Your Own Data
Making Data in Excel & R
Making Vectors of Data
Selecting & Filtering Data
Grouping & Summarizing Data
Pivoting Data in Excel & dplyr

40
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Module 7 - Multiple Tables of Data

Joining Data
Joining Data in Excel
Joining Data in R
Sorting Data
Hands-on Lab #3

Module 8 - Data Visualization

Introducing ggplot2
Box Plots
Histograms
Bar Charts
Scatter Plots
Hands-on Lab #4

41
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Days 2 & 3 - Introduction to Machine Learning With R

Module 9 - Supervised Learning
You Are the Teacher
What is Machine Learning?
Supervised Learning
Classification vs Regression
Why Decision Trees?
Ease of Use
Optimal for Many Business Problems

Module 10 - Classification Trees

Basic Intuition
Trees Are Rules
Sample Decision Tree
Overfitting Intuition
The Bugbear of Machine Learning
The Model Is Good! Or Is It?

Module 11 - Classification Tree Math

Gini Impurity
Gini Change
Many Categories Impurity
Numeric Feature Impurity

42
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Module 12 - Using Classification Trees

Classification Trees, tidymodels Style
The tidymodels Universe
Recipes
Model Specifications
Workflows
Model Fitting
Hands-on Lab #5

Module 13 - Introducing the Bias-Variance Tradeoff

Under/Overfitting
The Goldilocks Zone
Controlling Complexity
The Bias-Variance Tradeoff
Intuitive Example
Model Example

Module 14 - Model Tuning

Supervising the Data
Splitting the Data
Cross-Validation
Model Tuning Intuition
Making an Intuitive Example Real
Estimating Generalization Error
What About the Test Set?
Pruning Classification Trees
Pruning Intuition
Pre/Post-Pruning

43
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Module 15 - Model Tuning With tidymodels

Measuring Model Accuracy
Accuracy
Confusion Matrices
Sensitivity
Specificity
Model Tuning with Tidymodels
Setting up Cross-Validation
Cross-Validation Results
Tuning the Tree
Tuning the Results
Hands-on Lab #6

Module 16 - Feature Engineering

Intuition
What is Feature Engineering?
Extracting Features
Row vs Column Features
Data Leakage
What Is It?
An Example
tidymodels to the Rescue
Engineering Features for Decision Trees
Visualizing Decision Boundaries
Concepts to Remember
Missing Data
Why Is Data Missing
Dealing With Missing Data
Imputation With tidymodels
Hands-on Lab #7

44
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Module 17 - Regression Trees

The Basics
Regression Trees Minimize SSE
Calculating SSE
Numeric Feature SSE
Many Categories SSE
Regression Trees With tidymodels
Measuring "Accuracy"
Model Specification
Regression Trees in Practice

Module 18 - The Mighty Random Forest

Bad, Tree! Bad!
Decision Tree Variance
High Variance Leads to Overfitting
Real-World Decision Trees
Ensembles
Wisdom of the Crowd
Manufacturing Independence
Bagging
Randomizing Rows
Bagging in Action
The Power of Bagging
Feature Randomization
Intuition
Randomizing Columns
Feature Randomization in Action

45
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Module 19 - Using the Random Forest

Tuning Random Forests
The Bias-Variance Tradeoff
Random Forest Hyperparameters
Feature Importance
Out of Bag (OOB) Data
Permutation Importance
Random Forests With tidymodels
Hands-on Lab #8

Module 20 - Bootcamp Wrap-Up

Want to Kaggle?
Additional Resources

46
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

Ready to Empower Your Team?

As this document clearly demonstrates, if your team has Excel skills I can teach them R
programming. R programming opens up the universe of advanced analytics of which
machine learning is one example.

My speciality is customizing analytics training for your team to ensure your investment
generates business impact.

This customization takes on two primary forms:

1. Delivering a curriculum optimized for your team, either on-site or virtually.

2. Providing ongoing virtual coaching services to cement skills with your team.

Here are some common curriculum customizations I delivered to clients:

Delivering days 2 & 3 of the bootcamp to teams that already know R.

Adding 2 days of training on linear and logistic regression with R.
Swapping out machine learning for 2 days of linear/logistic regression with R.

Many of my clients augment bootcamp training with 1-on-1 coaching services for their
team. By coaching your team on how to apply the skills learned to your specific business
problems, your return on investment is ensured.

BTW - For my coaching clients I offer a guarantee. I will continue to coach your team free
of charge until they are successful with the learned skills.

Interested in learning more?

Email me: dave@daveondata.com

47
 FROM EXCEL TO MACHINE LEARNING BOOTCAMP

About the Author

My name is Dave Langer and I am the founder of Dave on

Data.

I'm a hands-on analytics professional, having used my skills

with Excel, SQL, and R to craft insights, advise leaders, and
shape company strategy.

I'm also a skilled educator, having trained 100s of working

professionals in a live classroom setting and 1000s more via
my online courses and tutorials.

In the past, I’ve held analytics leaderships roles at

Schedulicity, Data Science Dojo, and Microsoft.

I am proud to partner with TDWI to deliver the best

analytics training to ANY team.

Drop me an email if you have any questions:

dave@daveondata.com

48