Module 4 - Logistic Regression

Question 1

Logistic Regression?

Answer 1

Used for predicting the outcome of binary dependent variable

Question 2

Softmax Regression?

Answer 2

Multiclass Regression

Question 3

Linearly Separable?

Answer 3

A data set can be perfectly separated by a linear classifier

Question 4

Decision Boundary?

Answer 4

The surface separating different predicted classes

Question 5

Linear Classifier?

Answer 5

A classifier that learns linear decision boundaries

Question 6

Support Vector?

Answer 6

A training example (data point) not in the flat part of the loss diagram

A data point that is incorrectly classified or close to the boundary

Question 7

Smaller gamma leads to?

Answer 7

Smoother boundaries

Question 8

Why highest gamma is a bad option?

Answer 8

Overfitting

Question 9

The effectiveness of SVM depends on?

Answer 9

Selection of Kernel
Kernel Parameters
Soft Margin

Question 10

Selection of Kernel?

Answer 10

Gaussian kernel

Question 11

Kernel’s Parameters?

Answer 11

Gaussian kernel has a single parameter ^y^

Question 12

Soft margin parameter?

Answer 12

C

Question 13

Logistic Regression Comparison?

Answer 13

Is a linear classifier
Can use with kernels but slow
Outputs meaningful probabilities
Multi-class extension
All data points affect fit
L2 or L1 Regularization

Question 14

Support Vector Machines?

Answer 14

Linear classifier
Can use with kernels, fast
Does not naturally output probabilities
Multi-class extension
Only "support vectors" affect fit
Conventionally just L2 regularization

Question 15

Applications for SVM?

Answer 15

Reduces the need for labeled training instances
Helpful in text categorization
Image classification support
Widely applied in biological and other sciences

Question 16

Issues of SVM?

Answer 16

Requires full labeling of input data
Uncalibrated class membership probabilities
The SVM is only direct applicable for two-class tasks
Parameters of solved model are difficult to interpret