Data Analytics with Python

Introduction to Model Building

 Learning Objectives

By the end of this lesson, you will be able to:

Define machine learning

Explain the machine learning approach

List relevant terminologies that help you understand a dataset

Discuss the features of supervised and unsupervised learning

models

Explain algorithms such as regression, classification,

clustering, and dimensionality reduction
Introduction to Machine Learning
 Why Machine Learning?

If we stored the data generated in a day on Blu-ray disks and stacked them up, it would be equal to the height
of four Eiffel towers. Machine learning helps analyze this data easily and quickly.

Machine Learning
 Purpose of Machine Learning

Machine learning is a great tool to analyze data, find hidden data patterns and relationships, and extract
information to enable information-driven decisions and provide insights.

Identify patterns and

relationships

Data

Gain insights
into unknown
data

Take information-
driven decisions
 Machine Learning Terminologies

These are some machine learning terminologies that you will come across in this lesson:

Inputs Attributes

Features

Label Records

Response Observations

Outcome Examples

Target Samples
Machine Learning Approach
 Machine Learning Approach
The machine learning approach starts with either a problem that you need to solve or a given dataset that
you need to analyze.

Strive for
accuracy
Train and test
the model
Choose the right
model
Identify the
problem type
Extract the
features from
Understand the the dataset
problem/dataset
 Steps 1 and 2: Understand the Dataset and Extract Its Features

Let us look at a dataset and understand its features in terms of machine learning.
Features Response
(attributes) (label)
Education Professional Training Hourly Rate
(Yrs.) (Yes/No) (USD)
16 1 90
15 0 65
12 1 70
18 1 130
Observations
(records) 16 0 110
16 1 100
15 1 105
31 0 70

Predictors
 Steps 3 and 4: Identify the Problem Type and Learning Model

Machine learning can either be supervised or unsupervised. The problem type should be selected
based on the type of learning model.

Concept Problem Types Example

Supervised Learning Unsupervised Learning

• In supervised learning, the dataset used to train a • In unsupervised learning, the response or the
model should have observations, features, and outcome of the data is not known.
responses. The model is trained to predict the right
response for a given set of data points. • Unsupervised learning models are used to identify
and visualize patterns in data by grouping similar
• Supervised learning models are used to predict an types of data.
outcome.
• The goal of this model is to represent data in a way
• The goal of this model is to generalize a dataset so that meaningful information can be extracted.
that the general rule can be applied to new data as
well.
 Steps 3 and 4: Identify the Problem Type and Learning Model

Data can either be continuous or categorical. Based on whether it is supervised or unsupervised learning, the
problem type will differ.

Concept Problem Types Example

Supervised Learning Unsupervised Learning

Data Data

Data Type Data Type

Continuous Categorical Continuous Categorical

Problem Dimensionality Problem

Regression Classification Clustering
Type reduction Type
 Steps 3 and 4: Identify the Problem Type and Learning Model

Some examples of supervised and unsupervised learning models are:

Concept Problem Types Example

Supervised Learning Unsupervised Learning

Categories of news based on the topics Grouping of similar stories on different news networks
 Working of Supervised Learning Model

In supervised learning, a known dataset with observations, features, and response is used to create and
train a machine learning algorithm. A predictive model, built on top of this algorithm, is then used to predict
the response for a new dataset that has the same features.

New or
Known Data
Unseen Data

Observations/ Observations/
Records Records

Features/
Attributes
Predictive Features/
Model Attributes
Response/
Label Machine
Learning
Algorithm Predicted
Response/Label
 Working of Unsupervised Learning Model

In unsupervised learning, a known dataset has a set of observations with features, but the response is not
known. The predictive model uses these features to identify how to classify and represent the data points of
new or unseen data.

New or
Known Data
Unseen Data

Observations/ Observations/
Records Records

Machine Predictive
Features/ Features/
Learning Model
Attributes Attributes
Algorithm

Data
Representation
 Steps 5 and 6: Train, Test, and Optimize the Model

To train supervised learning models, data analysts usually divide a known dataset into
training and testing sets.

Supervised Learning Unsupervised Learning

Known Data Known Data

Observations/ Observations/
Records Records

Features/
Attributes
Features/
Attributes
Response/
Label
 Steps 5 and 6: Train, Test, and Optimize the Model

Known Data
Train
(60%-80%)

Test Observations/
(20%-40%) Records

Features/
Attributes

Response
/ Label Machine
Learning
Algorithm
 Steps 5 and 6: Train, Test, and Optimize the Model

Let us look at an example to see how the split approach works.

Model Training
Observation Response

ID Education Professional training Hourly rate

10 16 1 90
Train set Train set
45 15 0 65
Test set Test set
83 12 1 70

45 18 1 130

54 16 0 110

67 16 1 100

71 15 1 105

31 15 0 70
 Supervised Learning Model Considerations

Performance
Response optimization

Model Accuracy

Features
Generalization
Scikit-Learn
 Scikit-Learn

Scikit is a powerful and modern machine learning Python library for fully and semi-
automated data analysis and information extraction.

Efficient tools to identify Free and open Rich set of libraries

and organize problems datasets for learning and
(Supervised/Unsupervised) predicting

Model support for Model Open source

every problem type persistence community and
vendor support
 Scikit-Learn: Problem-Solution Approach

Scikit-learn helps data scientists organize their work through its problem-solution approach.

Model Estimator Model Model

Predictions Accuracy
Selection Object Training Tuning
 Scikit-Learn: Problem-Solution Considerations

While working with a Scikit-Learn dataset or loading your own data to Scikit-Learn, consider
these points:

✔ Create separate objects for feature and response

✔ Ensure that features and response have only numeric values

✔ Features and response should be in the form of a NumPy ndarray

✔ Since features and response would be in the form of arrays, they would have shapes and sizes

✔ Features are always mapped as x and response is mapped as y

Supervised Learning Models: Linear Regression
 Supervised Learning Models: Linear Regression

Linear regression is a supervised learning model used to analyze continuous data.

It is easy to use as the model does

not require a lot of tuning

It is the most basic and widely used

technique to predict a value of an
attribute

It runs very fast, which makes it

time-efficient
 Supervised Learning Models: Linear Regression

The linear regression equation is based on the formula for a simple linear equation.

Simple linear equation

Linear regression equation

Response Input features

Coefficient of x
Intercept
 Supervised Learning Models: Linear Regression

Linear regression is the most basic technique to predict a value of an attribute.

Data point

Residual
y Residual
(response) Least square line
dy

dx

(0, y) Slope/gradient
Actual Predicted
value value

x (predictor variable)

! The attributes are usually fitted using the least square approach.
 Supervised Learning Models: Linear Regression

Data point

y SSE
(response) Least square line
SSR

Regression of sum of squares

(0, y)

x (predictor variable)
Error of sum of squares

Smaller the value of SSR or SSE, the more accurate the prediction will be, which would make the
! model the best fit.
 Supervised Learning Models: Linear Regression

Let us see how linear regression works in scikit-learn.

Normalizes the regression

variable before performing
the regression operation

Calculates the intercept

Class for this model

sklearn.linear_model.LinearRegression(fit_intercept=True, normalize=False, copy_X=True, n_jobs=1)

Copies the Number of jobs to use

regression variable for the computation
Loading a Dataset

Problem Statement: Demonstrate how to load a built-in scikit-learn dataset.

Access: To execute the practice, follow these steps:
• Go to the PRACTICE LABS tab on your LMS
• Click the START LAB button
• Click the LAUNCH LAB button to start the lab
Linear Regression Model

Problem Statement: Demonstrate how to create and train a linear regression model.
Access: To execute the practice, follow these steps:
• Go to the PRACTICE LABS tab on your LMS
• Click the START LAB button
• Click the LAUNCH LAB button to start the lab
Supervised Learning Models: Logistic Regression
 Supervised Learning Models: Logistic Regression

Logistic regression is a generalization of the linear regression model used for classification problems.

Probability of y = 1, given x
Change in the log-
odds for a unit
change in x

The above equation is the simplest logistic function used for performing logistic regression.
 Supervised Learning Models: Logistic Regression

To interpret the outputs of a logistic function, you must understand the

difference between probability and odds.

Probability

Logarithm of odds Linear regression

 Supervised Learning Models: Logistic Regression

Inverse of
Specifies the norm used
regularization
in penalization
Calculates the intercept
Implemented only
Class for L2 penalty

class sklearn.linear_model.LogisticRegression(penalty='l2', dual=False, tol=0.0001, C=1.0, fit_intercept=True, intercept_scaling=1,

class_weight=None, random_state=None, solver='liblinear', max_iter=100, multi_class='ovr', verbose=0, warm_start=False, n_jobs=1)

Seed or the random Algorithm to use in the

If true, reuse the
state instance optimization problem
solution of the Number of
previous call jobs in parallel
Can be ovr (binary)
computation
or multinomial
Supervised Learning Models: K-Nearest Neighbors
 Supervised Learning Models: K-Nearest Neighbors (K-NN)

K-Nearest Neighbors, or K-NN, is one of the simplest machine learning algorithms used for both
classification and regression problem types.

Features
(Attributes)
 Supervised Learning Models: K-Nearest Neighbors

K=6
K=3

If you are using this method for binary classification, choose an odd number for k to avoid the case of a tied
distance between two classes.
 Supervised Learning Models: K-Nearest Neighbors
It looks at the inputs or features of the training dataset to identify the attributes of any new or unseen
data. Based on how similar a data point is to an attribute, the algorithm classifies it.

Features Response
(Attributes) (label)
K-NN and Logistic Regression Models

Problem Statement: Demonstrate the use of K-NN and logistic regression models.
Access: To execute the practice, follow these steps:
• Go to the PRACTICE LABS tab on your LMS
• Click the START LAB button
• Click the LAUNCH LAB button to start the lab
Unsupervised Learning Models: Clustering
 Unsupervised Learning Models: Clustering

A cluster is a group of similar data points.

Clustering is used to:

• Extract the structure of the data
• Identify groups in the data

Greater similarity between data points results in better clustering.

 Unsupervised Learning Models: K-Means Clustering
Scenario: You are given a dataset where each observed example has a set of features but has no labels or
response attached to it. So, in the absence of a response, you can identify which data points in a dataset
are similar. Each similar group of data points is called a cluster.
Centroids (mean)

Assign Optimize

Find the number of clusters and assign Iterate and optimize the mean for each cluster for
mean its respective data points
 Unsupervised Learning Models: K-Means Clustering
K-means finds the best centroids by alternatively assigning random centroids to a dataset and
selecting mean data points from the resulting clusters to form new centroids. It continues this
process iteratively until the model is optimized.

Assign data points to the centroids

Unsupervised Learning Models: K-Means Clustering

Choose a mean from each cluster as a centroid

Unsupervised Learning Models: K-Means Clustering

Reassign data points to new centroids

Unsupervised Learning Models: K-Means Clustering

Iterate the process till the model is optimized

 Unsupervised Learning Models: K-Means Clustering

Let us see how the K-means algorithm works in scikit-learn.

Number of clusters to form

and number of centroids to
generate Number of times the K-
means algorithm will be run
Precompute for
with different centroid seeds
Class faster operation
Selects initial cluster centers

sklearn.cluster.KMeans(n_clusters=8, init='k-means++', n_init=10, max_iter=300, tol=0.0001,

precompute_distances='auto',
verbose=0, random_state=None, copy_x=True, n_jobs=1)
Maximum number of
Initialize the iterations of the K-means
centers Number of jobs in algorithm for a single run
parallel computation
If true, does not
modify data
while
precomputing
K-Means Clustering to Classify Data Points

Problem Statement: Demonstrate how to use K-means clustering to classify data points.
Access: To execute the practice, follow these steps:
• Go to the PRACTICE LABS tab on your LMS
• Click the START LAB button
• Click the LAUNCH LAB button to start the lab
Unsupervised Learning Models: Dimensionality Reduction
 Unsupervised Learning Models: Dimensionality Reduction

It reduces a high-dimensional dataset into a dataset with fewer dimensions. This makes it easier and faster for the
algorithm to analyze the data.
 Unsupervised Learning Models: Dimensionality Reduction

These are some techniques used for dimensionality reduction:

Drop data columns with missing

values

Drop data columns with low

variance

Drop data columns with high

correlations
Large dataset
(a few thousand columns and
rows)
Apply statistical functions - PCA
Unsupervised Learning Models: Principal Component Analysis
 Unsupervised Learning Models: Principal Component Analysis (PCA)
It is a linear dimensionality reduction method which uses singular value decomposition of the data and
keeps only the most significant singular vectors to project the data to a lower dimensional space.

• It is primarily used to compress or reduce the

data.
• PCA tries to capture the variance which helps it
pick up interesting features. Minor
axis Principal axis
• PCA is used to reduce dimensionality in the
dataset and to build feature vector.
• Here, the principal axis in the feature space
represents the direction of maximum variance in
the data.

This method is used to capture variance.

Unsupervised Learning Models: Principal Component Analysis

Let us look at how the PCA algorithm works in scikit-learn.

Number of components to keep

Class

sklearn.decomposition.PCA(n_components=None, copy=True, white

n=False)

Overwrites the transform data after Removes data with

fitting them into the model lower variance
Principal Component Analysis (PCA)

Problem Statement: Demonstrate how to use the PCA model to reduce the dimensions of a dataset.
Access: To execute the practice, follow these steps:
• Go to the PRACTICE LABS tab on your LMS
• Click the START LAB button
• Click the LAUNCH LAB button to start the lab
 Pipeline

All models in the

pipeline must be Once all the data is
It simplifies the process
transformers. The last fit into the models
where more than one
model can either be a or estimators, the
model is required or
transformer or a predict method can
used.
classifier, regressor, or be called.
other such objects.

Estimators are known as model instance.

Build Pipelines

Problem Statement: Demonstrate how to build a pipeline.

Access: To execute the practice, follow these steps:
• Go to the PRACTICE LABS tab on your LMS
• Click the START LAB button
• Click the LAUNCH LAB button to start the lab
 Model Persistence

You can save your model for future use. This avoids the need to retrain the model.

• This can be saved using the Pickle method.

• It can also be replaced with the joblib of scikit team.
• Both joblib.dump and joblib.load can be used.
• These would be efficient for Big Data.
Persist a Model for Future Use

Problem Statement: Demonstrate how to persist a model for future use.

Access: To execute the practice, follow these steps:
• Go to the PRACTICE LABS tab on your LMS
• Click the START LAB button
• Click the LAUNCH LAB button to start the lab
 Model Evaluation: Metric Functions

You can use the metrics function to evaluate the accuracy of your model’s predictions.

metrics. accuracy_score
Classification metrics.average_precision_score

Clustering metrics.adjusted_rand_score

metrics.mean_absolute_error
Regression metrics.mean_squared_error
metrics.median_absolute_error
Project 1: Create a Model to Predict the Sales Outcome

Problem Statement:
The given dataset contains ad budgets for different media channels and
the corresponding ad sales of the firm. Evaluate the dataset to:
• Find the features or media channels used by the firm
• Find the sales figures for each channel
• Create a model to predict the sales outcome
• Split as training and testing datasets for the model
• Calculate the Mean Square Error (MSE)
Instructions to perform the assignment:
Download the FAA dataset from the “Resource” tab. Upload the dataset
to the JupyterLab to view and evaluate it.
Project 2: List the Glucose Level Readings

Problem Statement:
The given dataset lists the glucose level readings of several pregnant
women taken either during a survey examination or routine medical care.
It specifies if the two hours post-load plasma glucose was at least 200
mg/dl. Analyze the dataset to:
• Find the features of the dataset
• Find the response label of the dataset
• Create a model to predict the diabetes outcome
• Use training and testing datasets to train the model
• Check the accuracy of the model
Project 2: List the Glucose Level Readings

Instructions to perform the assignment:

• Download the “pima-Indians-diabetes.DATA” and “pima-Indians-

diabetes.NAMES” files from the “Resources” tab. Load the .DATA file to the
JupyterLab notebook to work on it.

• Open the .NAMES file with a notepad application to view its text. Use this
file to view the features of the dataset and add them manually in your
code.
 Key Takeaways

You are now able to:

Define machine learning

Explain the machine learning approach

List relevant terminologies that help you understand a dataset

Discuss the features of supervised and unsupervised learning

models
Explain algorithms such as regression, classification,
clustering, and dimensionality reduction
Knowledge Check
Knowledge
Check
In machine learning, which one of the following is an observation?
1

a. Features

b. Attributes

c. Records

d. Labels
 Knowledge
Check
In machine learning, which one of the following is an observation?
1

a. Features

b. Attributes

c. Records

d. Labels

The correct answer is c

An observation is a set of examples, records, or samples.

Knowledge
Check If data is continuous and has labels (response), then it fits which of the following problem
types?
2

a. Supervised learning: Classification

b. Unsupervised learning: Clustering

c. Unsupervised learning: Dimensionality reduction

d. Supervised learning: Regression

 Knowledge
Check If data is continuous and has labels (response), then it fits which of the following problem
types?
2

a. Supervised learning: Classification

b. Unsupervised learning: Clustering

c. Unsupervised learning: Dimensionality reduction

d. Supervised learning: Regression

The correct answer is d

The regression algorithm belonging to the supervised learning model is best suited to analyze continuous data.
Knowledge
Check
Identify the goal of unsupervised learning. Select all that apply.
3

a. To predict the outcome

b. To understand the structure of the data

c. To generalize the dataset

d. To represent the data

 Knowledge
Check
Identify the goal of unsupervised learning. Select all that apply.
3

a. To predict the outcome

b. To understand the structure of the data

c. To generalize the dataset

d. To represent the data

The correct answer is b, d

The goal of unsupervised learning is to understand the structure of the data and represent it. There is no right or
certain answer in unsupervised learning.
Knowledge
Check
The estimator instance in scikit-learn is a _____.
4

a. Model

b. Feature

c. Dataset

d. Response
 Knowledge
Check
The estimator instance in scikit-learn is a _____.
4

a. Model

b. Feature

c. Dataset

d. Response

The correct answer is a

The estimator instance or object is a model.

Knowledge
Check
What is the best way to train a model?
5

a. Use the entire dataset as both training and testing set

b. Split the known dataset into separate training and testing sets

c. Ask the source to provide continuous data

d. Ask the source to provide categorical data

 Knowledge
Check
What is the best way to train a model?
5

a. Use the entire dataset as both training and testing set

b. Split the known dataset into separate training and testing sets

c. Ask the source to provide continuous data

d. Ask the source to provide categorical data

The correct answer is b

The best way to train a model is to split the known dataset into training and testing sets. The testing set varies from
20% to 40%.
Knowledge
Check
Which of the following is true with a greater value of SSR or SSE? Select all that apply.
6

a. The prediction will be more accurate, making it the best fit model.

b. The prediction will start becoming less accurate.

c. The outcome remains unaffected.

d. The model will not be the best fit for the attributes.
 Knowledge
Check
Which of the following is true with a greater value of SSR or SSE? Select all that apply.
6

a. The prediction will be more accurate, making it the best fit model.

b. The prediction will start becoming less accurate.

c. The outcome remains unaffected.

d. The model will not be the best fit for the attributes.

The correct answer is b, d

With higher SSR or SSE, the prediction will be less accurate and the model will not be the best fit for the attributes.
Knowledge
Check
Class sklearn.linear_model.LogisticRegression, random_state _____.
7

a. Indicates the seed of the pseudo random number generator used to shuffle data

b. Defines the features state

c. Represents the number of random iterations

d. Specifies a random constant to be added to the decision function

 Knowledge
Check
Class sklearn.linear_model.LogisticRegression, random_state _____.
7

a. Indicates the seed of the pseudo random number generator used to shuffle data

b. Defines the features state

c. Represents the number of random iterations

d. Specifies a random constant to be added to the decision function

The correct answer is a

The class “sklearn.linear_model.LogisticRegression, random_state” indicates the seed of the pseudo random number
generator used to shuffle data.
Knowledge
Check
What are the requirements of the K-means algorithm? Select all that apply.
8

a. Number of clusters should be specified

b. More than one iteration should meet requisite criteria

c. Centroids should minimize inertia

d. Features should be labeled

 Knowledge
Check
What are the requirements of the K-means algorithm? Select all that apply.
8

a. Number of clusters should be specified

b. More than one iteration should meet requisite criteria

c. Centroids should minimize inertia

d. Features should be labeled

The correct answer is a, b, c

The K-means algorithm requires the number of clusters to be specified and the centroids to minimize inertia. It
requires several iterations to fine tune itself and meet the required criteria to become the best fit model.
Knowledge
Check In Class sklearn.decomposition.PCA, the transform(X) method, where X is multi-dimensional,
_____.
9

a. Fits the model with X and applies the dimensionality reduction on X

b. Transforms the data back to its original space

c. Applies the dimensionality reduction on X

d. Computes data co-variance with the generative model

 Knowledge
Check In Class sklearn.decomposition.PCA, the transform(X) method, where X is multi-dimensional,
_____.
9

a. Fits the model with X and applies the dimensionality reduction on X

b. Transforms the data back to its original space

c. Applies the dimensionality reduction on X

d. Computes data co-variance with the generative model

The correct answer is c

In Class “sklearn.decomposition.PCA,” the transform(X) method applies the dimensionality reduction on X.

Thank You