Counterpoint Reinforcement Learning

An agent for composing species counterpoint.

Dependencies

Python dependencies are given in requirements.txt. TensorFlow must be installed to run the deep learning experiments. Lillypond, for typesetting compositions. Timidity for MIDI playback and conversion.

Usage

Execute runner.py for a usage message. The experiments in the report were run simply by specifying the task (0) and the approach (see the file for the list of these).

Interesting files to look at include [species_one.py](counterpoint/species/species_one.py) where you'll find the reward function, and [runner.py](runner.py) where you'll find the list of experiments and the CLI.

Attribution

CMAC implementation is provided by Jeremy Stober.

Reinforcement learning classes reused from my previous programming assignments.

Substantial musical analysis functionality is provided by Abjad.