bozicb/similarity-theory

similarity-theory

This repository consists of Ruby scripts and functions for similarity theory. I'm not making a claim of mathematical identity, but it might be a useful set of helper functions for your own machine learning project, especially if you working with students and plan to look under the hood of similarity models.

In the following, I am giving an overview of Ruby scripts and functions for classification tasks in similarity theory. Every module covers one part which is used in our machine learning classes.

knn.rb

The k-NN module covers the k-Nearest Neighbours model with several similarity metrics.

knn(DataFrame, int, Array, (opt)String)

The knn function takes a Daru::DataFrame object which has the following structure:

Cont. Feature 1	Cont. Feature 2	...	Cont. Feature n	Target Feature
random value	random value	...	random value	random value/category
...	...	...	...	...

The next parameter is the amount of neighbours you want to take into account, so that's our k. Then we have the query Array that needs to have the same amount of continuous features as our dataset, but no target. The last parameter is optional. It selects the similarity metric to be used.

The return value of this function will be a prediction for the query.

normalise(Daru::Vector)

Normalises the values of a Daru::Vector.

euclidean(Array, Array)

This function uses instances a and b and returns the euclidean distance between them. As in the following equation: