MLE

• Use log likelihood formula argmax log P(Y | X)
• derive w/ respect to X, set = 0

Perceptrons

Update Rule

• w
• We subtract when y_true = -1 and y_pred = 1

Properties

Naive Bayes & Smoothing

MLE for Naive Bayes

• We use log properties to convert the product to a sum

  

• then we set it to 0 and again solve for our estimation parameter, giving us the maximizer

Laplace Smoothing

• large k: reduce overfitting