Practice

Abstract

In this chapter, we will explain how to install an experimental environment based on Python, and present a few examples of basic Python operations and the code for the visualisation of a simple dataset (as in Example 1.4) and the confusion matrix (as introduced in Sect. 2.4).

Exercises

Question 1

Give an example for a supervised learning problem, an unsupervised learning problem, and a semi-supervised learning problem; □

Question 2

Give an example of a supervised learning problem that is a classification task; □

Question 3

Give an example of a supervised learning problem that is a regression task; □

Question 4

Give an example of a clustering problem; □

Question 5

For all the above problems, figure out the features and labels for them; □

Question 6

Write a program to output the following information:

• 1. How many samples are in the iris dataset;

• 2. How many features are given for each sample in the iris dataset?

• 3. What is the value range for each feature?

□

Question 7

According to Table 4.1 about two random variables Intelligence and Grade, please compute the values P(Grade = B | Intelligence = Low) and MAP(Grade).

Table 4.1 Joint probability for student grade and intelligence

Question 8

Consider a joint distribution table as in Table 4.2, can you compute the following expressions:

Table 4.2 Joint distribution table

• P(A=1)=0.6

• P(A=2)=0.3

• P(B=3)=0.4

• P(B=4)=0.1

• P(A=1—B=2)=0.6

• P(B=3—A=3)=0.4

• MAP(A—B=2)=1

• MAP(B—A=2)=3

• MAP(A)=1

• MAP(B)=3

□

Question 9

Consider a joint distribution table as in Table 4.3, can you compute the following expressions:

Table 4.3 Joint distribution table

• P(A=1)=0.56

• P(A=2)=0.3

• P(B=3)=0.4

• P(B=4)=0.06

• P(A=1—B=2)=0.6

• P(B=3—A=3)=2/7

• MAP(A—B=2)=1

• MAP(B—A=2)=3

• MAP(A)=1

• MAP(B)=3

□

Question 10

Write a program to implement

• ROC curve

• PR curve

□

Question 11

Compare a few training/test splits (0.9/0.1, 0.8/0.2, 0.7/0.3) and check their differences on training and test accuracy. □

Question 12

Compare a few training/test splits (0.9/0.1, 0.8/0.2, 0.7/0.3) and check their differences on confusion matrix. □

Question 13

Find a data poisoning strategy to make the trained model mis-classify on a given training input. □

Question 14

Read the literature to understand the state-of-the-art for backdoor attacks. □