Closed satpreetsingh closed 5 months ago
Hey! This was a simplification we made when collecting baselines, it is fixed in this file by adding config["GRU_HIDDEN_DIM"]
. We are looking to update the rest of the scripts over the next month.
fixed with #77
Hi, why are the RNN HIDDEN_SIZE and NUM_STEPS tied to 128? Other values seem to throw tensor reshaping errors.
Tried on IPPO and MAPPO with MPE, e.g. with NUM_STEPS=256:
I don't immediately see a reason from the algorithm why these should be tied up.