#laplace-smoothing

Question 1: Single choice

A naive Bayes digit classifier on binarized 784-pixel images is trained without Laplace smoothing. In training, pixel 17 was never on in any image of class 3. A test image has pixel 17 on. What score $P(y = 3)\prod_i P(x_i \mid y = 3)$ does class 3 get?

Zero (log-score $-\infty$ ), whatever the other 783 pixels say1
A slightly lower score, since one unusual pixel is outvoted by the other 7832
Just the prior $P(y = 3)$ , because unseen pixel values are skipped3
The same score as with smoothing, since smoothing only changes the priors4

Timed test