AIDAID202 · Lectures 7–8

Lecture 8 · Naïve Bayes

Why the Naïve Bayes assumption helps

Conditional independence turns one enormous joint distribution into a product of small feature-wise likelihoods. That is both its strength and its potential weakness.

Predict first: With 20 binary features and 2 classes, how many class-conditional parameters would a full joint model need compared with Naïve Bayes?

Parameter explosion

Full joint model—
Naïve Bayes—

Build a posterior feature-by-feature

P(Enjoyable | x)—
P(Not enjoyable | x)—
Features are treated as conditionally independent given the class.