ScalaMI

This is ScalaMI, an implementation of MIToolbox in Scala.

It provides a series of functions for working with information theory. It also contains some variable manipulation functions to preprocess discrete/categorical variables to generate information theoretic values from the variables.

These functions are targeted for use with feature selection algorithms rather than communication channels and so expect all the data to be available before execution and sample their own probability distributions from the data.

Functions contained:

• Entropy
• Conditional Entropy
• Mutual Information
• Conditional Mutual Information
• generating a joint variable
• generating a probability distribution from a discrete random variable

Installation

Scala 2.11.x, 2.12.x and 2.13.x is supported.

You can either just clone this code or install package from GitHub's maven repo.

To install from Github packages (maven):

• use SBT Github Packages plugin
• install the plugin as specified on its repo. Then just this to your build.sbt:

resolvers += Resolver.githubPackages("konradmalik", "ScalaMI")

• after the resolvers are provided, package installation is easy:

libraryDependencies += "konradmalik" %% "scala-mutual-information" % "0.0.2"

• IMPORTANT: currently, there is a limitation on github's maven repo - only authorized users can download packages. What This means is you will need to have a PAT (Personal Access Token) available in GITHUB_TOKEN env variable at the time when you execute anything related to sbt. If your PAT is invalid, then any attempt on package installation will fail with status 400 or 401. Managing this env variable is a separate issue, but I recommend direnv

Another installation method is to just download the jar file from packages (you do not need to be authenticated) and manually include it as a dependency in your project.

Exemplary usage:

import konradmalik.ScalaMI

val entropy = ScalaMI.Entropy.calculateEntropy(Array(2.3, 4.1, 54.3))

TODO

• Tests
• Actions

Credits

This package is a rewrite of the JavaMI (https://github.com/Craigacp/JavaMI) from Java to Scala. Credits to Craigacp (JavaMI) and MIToolbox authors. New features might be added in the future, independently of JavaMI.