Description

Objective Question

Consider a K class classification problem. Total number of binary classifier trained in a one-vs-all setting are    , and for one-vs-one classifier setting are              .

2           Objective Question

Which of the following methods do we use to best fit the data in Logistic Regression?

1. Least Square Error
2. Maximum Likelihood
3. Jaccard Distance
4. Both A and B

3           Programming Question

In tutorial, we saw how to build a multi-class classifier using a one-vs-all setting. In this problem you are required to construct a one-vs-one classifier for the same problem. The starter code is provided in ‘Logistic Regression Excercise 1.ipynb’.

4           Subjective Question

Suppose you train a logistic regression classifier and your hypothesis function h is

h_θ(x) = g(θ₀ + θ₁x₁ + θ₂x₂)

where,

θ₀ = 6, θ₁ = 0, θ₂ = −1

Draw the decision boundary for the given classifier (a rough sketch is sufficient). What would happen to the decision boundary if you replace the coefficient of x₁ and x₂? Draw the decision boundary for the second case as well.

1

5           Programming Question

You have been given a dataset of handwritten digits, the MNIST dataset. The dataset consists of 28×28 pixel images consisting of one of 10 digits (0,1,…,9) that are handwritten. Using logistic regression in one-vs-one and one-vs-all setting construct a 10-class classifier. Report the test accuracy for one-vs-one and one-vs-all classifier. The starter code is provided in ‘Logistic Regression Excercise 2.ipynb’.

2