davpinto/fastknn: Build Fast k-Nearest Neighbor Classifiers

A fast KNN learner for binary and multinomial classification problems, build upon the ANN library. It has been developed to deal with very large datasets (> 100k rows). The 'fastknn' makes it easy to find the best 'k' and to plot beautiful decision boundaries for the classifiers. Moreover, it provides estimators for the class membership probabilities based on voting and weighted voting rules. The last one gives more calibrated probabilities in general, and reduces log-loss. If you want to combine KNN with other classifiers, 'fastknn' provides a feature extraction method that makes a nonlinear mapping from the original features using KNN. Then the new features can be used to improve the performance of any other classifier of your choice.

Getting started

Package details

URL https://github.com/davpinto/fastknn
Package repositoryView on GitHub
Installation Install the latest version of this package by entering the following in R:
davpinto/fastknn documentation built on May 15, 2019, 1:18 a.m.