COMS W4771 Lecture Notes - Lecture 4: Latex, Dynamic Programming, C4.5 Algorithm

• School: Columbia University
• Department: Computer Science
• Course: COMS W4771
• Professor: Germán Creamer

Document Summary

Looking at the labels of the closest observations and deriving a new label for a new example: where k stands for the # of closest labels looked at, cannot be an even number e. g. 5-nearest neighbors, e. g. In general for size of k: smaller k means smaller training error. Learning algorithm is consistent if sample size n]=( ) lim. Decision trees: decision tree is a function : represented by a binary tree. If yes then go down one direction, no for the other direction. ^ : gini index (used in economics, () : = 2(1 ) Showing economic disrepencies: entropy (from information theory) d e. Example: entropy = (pr = log d rs tuv, total of 30 observations, 14 with label a, and 16 with label b, calculating entropy: dz dw dw xy dx log log xy dx d} d} 0. 787 + xy w d} dx d} log log dz xy w d}