Reinforcement Learning Examples

These examples illustrate how to implement a couple reinforcement learning algorithms to play Atari games. This uses the OpenAI Gym through its Python api so the gym Python package has to be installed. The Python library is executed in Rust using rust-cpython.

The policy gradient example uses the CartPole-v0 environment. It can be run using the following command.

cargo run --example reinforcement-learning  --features=python pg

The A2C example can use any Atari game supported in the OpenAI gym. Training can be launched by running:

cargo run --example reinforcement-learning  --features=python a2c

This produces some weight files in the current directory.

If you get some Python import error when using conda, this can be related to rust-cpython not being able to use the Python runtime as a shared library. rust-cpython uses Py_ENABLE_SHARED to determine the link mode which does not seem to work properly with conda as reported in this issue.

A2C Agent Playing Breakout

A2C Agent Playing SpaceInvaders