genrl
genrl copied to clipboard
SforAiDl
Categorical DQN not training
There might be some shape related errors or we're missing something. Either that or hyperparameters need to be tuned.
One issue is that the shapes are inferred directly using foo.reshape(-1, some_size) instead of foo.reshape(some_known_size, some_size). Might take some playing around to figure out the problem.
The categorical_q_target function seems to be the main problem here. Tried to reshape and modify it a lot but didn't work. I'm guessing the problem is because of both n_envs and batch_size there seems to be a reshaping problem somewhere. It is not very easy to figure out which dimension refers to which variable while reshaping in the function.
Yes
On Sat, 29 Aug 2020, 09:33 Ajay Subramanian, [email protected] wrote:
Is this still an issue?
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/SforAiDl/genrl/issues/208#issuecomment-683232280, or unsubscribe https://github.com/notifications/unsubscribe-auth/AH72FJ57LIGMPZEDOOJLENTSDB4YLANCNFSM4PQBXD7Q .
