Bug fix for SAC-discrete.

[toshikwa]

Hi, thank you for your great work :)

I fixed some bugs related to #54 and #56.
I tested it in CartPole.py and saw the training converged more stably.

Changes

• fix min_qf_next_target to properly calculate expectations over policy.
• similarly fix policy_loss.
• fix max_probability_action to properly get argmax over each sample, not entire batch. (It doesn't actually affect the algorithm, though.)
• fix device for Replay_Memory to train SAC-Discrete on CPU.
• fix errors due to the update of PyTorch (now torch>=1.4.0 works!!).

Thanks :)

[p-christ]

thanks a lot

[toshikwa]

Hi @p-christ

Thank you for merging.
Could you close relevant issues(54, 56)??

Also, I found that this repo got older and there were a few bugs.
Can I fix bugs and clean up codes??

Thanks :)

[p-christ]

Sure will close them. Yes pls do clean up the code and do a new pull request!

…

On Sat, Sep 19, 2020 at 21:58, Toshiki Watanabe ***@***.***> wrote: Hi @p-christ <https://github.com/p-christ> Thank you for merging. Could you close relevant issues(54, 56)?? Also, I found that this repo got older and there were a few bugs. Can I fix bugs and clean up codes?? Thanks :) — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#60 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGJAGA6AU2JDIJ3WD7WFR6TSGULPJANCNFSM4RRFGROA> .