JRuby Mahout

Jruby Mahout is a gem that unleashes the power of Apache Mahout in the world of JRuby. Mahout is a superior machine learning library written in Java. It deals with recommendations, clustering and classification machine learning problems at scale. Until now it was difficult to use it in Ruby projects. You'd have to implement Java interfaces in JRuby yourself, which is not quick especially if you just started exploring the world of machine learning.

The goal of this library is to make machine learning at scale in JRuby projects simple.

This is an early version of a JRuby gem that only supports Mahout recommendations. It also includes a simple Postgres manager that can be used to manage appropriate recommendations tables. Unfortunately it's impossible to use ActiveRecord (AR) with Mahout, because AR operates at a much higher level and creates a lot of overhead that is critical when dealing with millions of records in real time.

First of all you need to download the Mahout library from one of the mirrors. Jruby Mahout only supports Mahout 0.7 at this point.

If you wish to work with a database for recommendations, you'll have to install the JDBC driver for Postgres. Another option is to use file-based recommendations.

platform :jruby do
  gem "jruby_mahout"
end

I am planning to add more examples covering Jruby Mahout use cases to this repo soon.

First, define the MAHOUT_DIR environmental variable for your Mahout installation. For example:

export MAHOUT_DIR=/bin/mahout

The easiest way to start working with Jruby Mahout recommendations is to initialize a recommender:

require 'jruby_mahout'
recommender = JrubyMahout::Recommender.new("PearsonCorrelationSimilarity", 5, "GenericUserBasedRecommender", false)

Set up a data model:

recommender.data_model = JrubyMahout::DataModel.new("file", { :file_path => "recommender_data.csv" }).data_model

and get recommendations:

puts recommender.recommend(2, 10, nil) # 10 recommendations for user with id = 2

You can evaluate your recommender to see how efficient it is:

puts recommender.evaluate(0.7, 0.3)

The closer the score is to zero—the better.

I am working on a series of articles on how to utilize JRuby Mahout in the real world projects. This is the first one in the series:

• Machine Learning with Ruby, Part One

There are several things that should be supported by this gem, before it can be used in production. Some of them are:

• Hadoop integration
• Clustering support
• Classification support
• Better docs

If you feel like you can help—please do.

Jruby Mahout is thoroughly tested with Rspec.

• Fork the project.
• Write code for a feature or bug fix.
• Add Rspec tests for it.
• Commit, do not make changes to rakefile or version.
• Submit a pull request.