SVM

Question 1

Maximum Margin Classifier/Separator

Answer 1

The decision boundary that maximizes separation boundary between separable classes

Question 2

Soft Margin

Answer 2

In SVC, the amount of error allowed when determining the width of the ‘street’.

The parameter is controlled by C in sklearn

Question 3

SVC Kernels

Answer 3

How the SVC/SVM transforms the data to create the hyperplane for separating the data

Question 4

Linear Kernel

Answer 4

No data transformation to determine the decision boundary in SVM/SVC

Question 5

Non linear kernel

Answer 5

Transformation of the data into a higher number of dimensions to determine the hyperplane that separates the classes

Question 6

When do you use non linear kernel in SVM

Answer 6

When the data is intermediate (either in number of samples or number of features). When features or samples are large, then the non linear kernel may be predictive but not very performant.

Question 7

Polynomial Kernel

Answer 7

Polynomial transformation of the data into higher number of dimensions in SVM

Question 8

Radial Basis Function (RBF) Kernel

Answer 8

Closest observations for a particular datapoint has a lot more weight to classify the current datapoint compared to observations that are further away.

Question 9

Gamma Parameter (SVM)

Answer 9

In RBF Kernel, this parameter scales the squared distances between two datapoints to determine how much influence nearest datapoints have on the classification of current point.

Higher gamma = More influence the two datapoints have on each other

Question 10

Support Vectors

Answer 10

Data points closest to the decision boundary that have the greatest influence on the position of the hyperplane

Question 11

SVM in 1 Sentence

Answer 11

A classification method that uses a hyperplane on higher dimensional transformed data to separate classes