Purity, splits, forests and boosting -- the family that still wins on tabular data.
8h · 11 lessons · 1 challenge
Before this: Applying Machine Learning
A decision tree asks one question at a time, and the whole subject is which question. This module derives that from purity and information gain, then grows the tree, then handles the awkward cases: categorical features, continuous features, and regression at the leaves.
Then the ensembles. Bagging and random forests reduce variance, boosting reduces bias, and gradient boosting still beats deep learning on most tables you will meet.
Grow a tree by hand, one split at a time, against the algorithm. Scored on how close your tree gets to its purity.
Worth 400 XP.
