Discovering reinforcement learning algorithms

A jax/stax implementation of the NeurIPS 2020 paper: Discovering reinforcement learning algorithms [1]

The agent at lpg.agent.py implements the bsuite.baseline.base.Agent interface. The lpg/environments/*.py interfaces with a dm_env.Environment. We wrap the gym-atari suite using the bsuite.utils.gym_wrapper.DMEnvFromGym adapter into a dqn.AtariEnv to implement historical observations and actions repeat.

Installation

To run the algorithm on a GPU, I suggest to install the gpu version of jax [4]. You can then install this repo using Anaconda python and pip.

conda env create -n lpg
conda activate lpg
pip install git+https://github.com/epignatelli/discovering-reinforcement-learning-algorithms

Note from Student

Pip installing the github link from above will install all requirements you need to run my student.ipynb.

1. To run my experiments you only need to run the code cells in the student.ipynb after pip installing said repo. (Depending on the device the visualizations/rendering might not work. It only works for one of my workstations)
2. a) Source code (all *.py files) is from https://github.com/epignatelli/discovering-reinforcement-learning-algorithms b) I've not touched the source code, only read it and applied it. c) As stated at the top of the student.ipynb, every piece of code that I wrote is in that file.
3. As this was a RL agent, I didn't use "datasets" on it, but instead had different environments, extensively described in my paper.

References

[1] Oh, J., Hessel, M., Czarnecki, W.M., Xu, Z., van Hasselt, H.P., Singh, S. and Silver, D., 2020. Discovering reinforcement learning algorithms. Advances in Neural Information Processing Systems, 33.