KnnCv

A native kNN leave-one-out technique implementation (oriented to feature selection) for Ruby based on the 'class' package for R.

Add this line to your application's Gemfile:

gem 'knn_cv'

And then execute:

$ bundle

Or install it yourself as:

$ gem install knn_cv

This gem provides very basic functionality in a straight-forward manner. Classifier objects can be created in order to evaluate kNN several times:

knn = KnnCv::Classifier.new k, instances, classes, numeric_attrs, Random.new

where k represents the number of neighbors to be considered, instances must be an Array of instances, which in turn are numeric Arrays, a classes Array with classes numbered from 0, and numeric_attrs a binary (0 or 1) vector indicating which attributes are to be treated as numeric. The last argument is a Random object that can be used to ensure reproducibility.

To evaluate the behavior of the classifier over certain features, use the fitness_for method:

knn.fitness_for [0, 1, 0, 0, 1, 0, 1, 1, 1]

The argument must be a binary Array (or other type of object that converts to a binary Array, such as BitArray).

Bug reports and pull requests are welcome at https://github.com/fdavidcl/knn_cv.