mirror of
https://github.com/wassname/DeepRL.git
synced 2026-08-21 11:09:46 +08:00
A really minor change. Should we input logits instead of probs into `Categorical`? It will do less converting from logits and back, leading to increased stability during training (theoretically).