# Homework 12 Implement the FSA variable selection

\$30.00

Rate this product

Homework 12
1. Implement the FSA variable selection method with linear models for multi-class
classification with the Vapnik loss:
L(u, y) = X
k6=y
`(uy − uk), (1)
where `(u) is the logistic loss described in class. Use the parameters λ = 0.0001, µ =
100, Niter = 500.
Take special care to normalize each column of the X matrix to have zero mean
and variance 1 and to use for normalizing the test set the same mean and standard
deviation that you used for normalizing the training set.
Assuming that the coefficient vector is a p × c matrix W, where p is the number of
features and c is the number of classes, use the norm kwjk, j = 1, …, p of each row as
the criterion to select the variables in FSA.
a) Using the satimage data, train a multi-class FSA classifier on the training set,
starting with β
(0) = 0 to select k ∈ {5, 9, 18, 27, 36} features. For each k find
an appropriate learning rate η to obtain a small final loss value on the training
set. Plot the training loss vs iteration number for k = 27. (5 points)
b) Report in a table the misclassification errors on the training and test set for the
models obtained for all these k. Plot the misclassification error on the training
and test set vs k. (1 point)
c) Repeat points a) and b) using the cross-entropy loss function:
L(u, y) = −uy + ln(X
C
k=1
exp(uk)), (2)
adding the misclassification errors to the table from b). (4 points).
1

Scroll to Top