Decision boundaries, the logistic link, and a cost function that comes from the likelihood.
7h · 9 lessons · 1 challenge
Before this: Supervised Learning & Linear Regression
Linear regression fails at classification, and the way it fails is instructive. This module replaces it: the logistic function, the decision boundary it draws, and the cost function that makes it trainable.
That cost is cross-entropy, which the probability module already derived from maximum likelihood. Here it arrives as the thing gradient descent actually minimizes.
A classifier and a cost matrix. Move the threshold to the point that costs the business least, then defend it.
Worth 350 XP.
