Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: years later (Friedman et al. 2000): –  Adaboost minimizes exponen%al loss func%on. •  There s%ll are open ques%ons. Random Forest •  Builds upon the idea of bagging •  Each tree build from bootstrap sample •  Node splits calculated from random feature subsets h4p://www.andrewbun%ne.com/ar%cles/about/fun Random Forest •  All trees are fully grown •  No pruning •  Two parameters –  Number of trees –  Number of features Random Forest Error Rate •  Error depends on: –  Correla%on between trees (higher is worse) –  Strength of single trees (higher is be4er) •  Increasing number of features for each split: –  Increases correla%on –  Increases strength of single trees Out of Bag Error •  Each tree is trained on a bootstrapped sample •  About 1/3 of data points not used for training •  Predict unseen points with each tree •  Measure error Out of Bag Error data points...
View Full Document

This document was uploaded on 12/22/2013.

Ask a homework question - tutors are online