Skip to content

Conversation

@natinew77-creator
Copy link

This PR includes three key improvements:

  1. Relaxed Dependency Constraints (Issue Notebook opened in colab cannot run successfully #347): Modified setup.py to use inclusive versioning (>=) for major packages like tensorflow, jax, and dm-reverb. This resolves installation issues in environments like Google Colab where pinned versions conflict with pre-installed packages. Also constrained gym<0.26.0 to ensure API compatibility.
  2. EpisodeAdder Documentation (Issue Not enough documentation for EpisodeAdder  #38): Added a detailed docstring to acme.adders.reverb.episode.EpisodeAdder explaining its purpose, arguments, and behavior.
  3. Decaying Epsilon Support: Enhanced acme.tf.networks.legal_actions.EpsilonGreedy to accept and use a shared tf.Variable for epsilon. This allows external schedules (e.g., decay) to update the exploration rate dynamically, addressing a previous FIXME.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant