Fickle

Experimental machine learning REST API.

Summary

This package contains modules for building a machine learning or recommender system using scikit-learn and python-recsys for use in a Flask WSGI application.

Configuration

All configuration is done using environment variables. The following variables are recognized:

• PORT: Web server port. Defaults to 5000.
• FICKLE_MODEL: Predictor or recommender model to use. Defaults to GenericSVMClassifier.
• FICKLE_DEBUG: Web server debug mode. Defaults to false.
• FICKLE_PASSWORD: Basic authentication password. Authentication is disabled when empty.
• REDIS_URL: Storage layer connection URL. Defaults to local instance.

Requirements

See requirements.txt for details.

Interface

See fickle-ruby for details.

Deployment

Heroku

$ git clone https://github.com/norbert/fickle && cd fickle
$ heroku apps:create -b git://github.com/norbert/heroku-buildpack-python.git#numpy
...
$ git push heroku master
...
$ heroku run ipython