Classification

Question 1

Classification

Answer 1

Regression, compute model using known data that will perform well on unknown data

Question 2

Evaluation

Answer 2

Accuracy
Speed
Robustness
Scalability
Interpretability

Question 3

Logistic Regression

Answer 3

Score(xi) = sum(wj*hj(xi))
yi = sign(Score(xi))
No compute the label, computes probability of assigning a class
P(yi = +1|xi) = 1/(1+e^(-score(xi)))
Search for the weight vector highest likelihood

Question 4

Overfitting and Regularization

Answer 4

L1 & L2
L1: sum of absolute values l(w)-alpha||w||_1
L2; sum of squares l(w)-alpha||w||_2^2

Question 5

K-fold crossvalidation

Answer 5

Generates k separate models
Evaluation on model performance on unknown data
Output its the evaluation, not the model

Question 6

Multiclass Classification

Answer 6

Logistic works with two class values
One versus the rest.

Question 7

Categorical Attributes

Answer 7

One Hot encoding

Question 8

Support Vector Machines

Answer 8

w^T + b = 0
Hyperplane that maximizes the maargin or largest y
yi(w*xi+b) >

Question 9

Neural Networks

Answer 9

wi*xi - b
Perceptron: value of weighted sum and returns its Sign
Basically a linear classifier, decision is hyperplane
Activation function: linear (regression), sigmoid, tanh (classification)