#wrapper-methods

Question 1: Single choice

Eight training examples $(x_1, x_2, x_3)$ :

label 0	label 1
$(3, 3, 0)$	$(2, 2, 0)$
$(3, 3, 1)$	$(1, 1, 1)$
$(3, 3, 0)$	$(1, 1, 0)$
$(2, 2, 1)$	$(1, 1, 1)$

A filter method that ranks single features by their mutual information with the label picks $x_1$ and $x_2$ . Which two features would a greedy wrapper (forward selection by the training accuracy of a linear logistic regression) end up with?

$x_1$ and $x_2$ , the same as the filter1
One of $x_1$ and $x_2$ , plus $x_3$ 2
$x_3$ first, then one of $x_1$ and $x_2$ 3
All three, because no pair reaches zero training error4

Timed test