Please report reproducible bugs through the issue tracker (please include a level_seed if applicable), as well as issues with installing or running on any supported platform.
Because these environments are meant for experiments, we would like to minimize changes. Changing anything that impacts training results will make it hard to compare results from previous experiments and so this should be avoided. As a result, PRs that add new functionality are unlikely to be accepted.
Feel free to fork this repo and make your own customizations. The build setup uses Travis to make creating a fork and producing your own packages relatively easy.