Logistic Regression

Question 1

What is logistic regression used for?

Answer 1

Prediction and classification models

Question 2

Binary Classification

Answer 2

Embedding a data point with label y E {0, 1} into a feature space f(x) E R^n. Determining in data point is positive or negative.

Question 3

Linear Binary Classification

Answer 3

Line specifies separation between positive and negative labels. Goal is to learn weight ‘w’ from data (wTf(x) > 0)

Question 4

BOW Features

Answer 4

Bag-of-words feature representation. Predefined vector where each index represents a different feature and its presence with a 0 or 1

Question 5

BOW and Classification

Answer 5

Can use BOW features for classification of a corpus.

Question 6

Weight Function

Answer 6

w = (w1, …, wn)T

• w is weight
• set to vector of scalars

Question 7

Data Point Function

Answer 7

x = (x1, …, xn)T (n features)

• x is data point
• set to vector of features

Question 8

Probability of Calculating Positive Weight Function

Answer 8

P(y = +|x) = logistic(wTx)

Question 9

How do you learn weight ‘w’?

Answer 9

Maximize log likelihood on data