AIDAID202 · Lectures 9–10

Lecture 9 · Modeling choices

Generative vs. discriminative learning

Gaussian Naïve Bayes models how features arise within each class. Logistic regression models the class probability directly.

Predict first: With little data and a roughly correct Gaussian/independence assumption, which model might stabilize faster? What if the assumption is badly wrong?

Same sample, two learners

Logistic boundaryGNB boundary
LR test accuracy—
GNB test accuracy—
n training—
Regime—