search

juliusfrost / dreamer-pytorch

Dream to Control: Learning Behaviors by Latent Imagination, implemented in PyTorch.
MIT License
282 stars 35 forks source link

Very minor bug in RSSM #81

Open alec-tschantz opened 3 years ago

alec-tschantz commented 3 years ago

Very minor (and I may have misunderstood): I believe the input to the stochastic_prior_model in the RSSMTransition class should be deterministic_size rather than hidden_size (line 76 in rnns.py).

This doesn't cause any errors at the moment as these default to the same value (200). But I see no reason why these have to be the same - one is the hidden size of the RNN, and the other is a generic size for hidden nodes in the model.

juliusfrost commented 3 years ago

Thanks for finding this! If you want, you can make a pull request with the small change!