Cannot reproduce the results for MPE

PKU-MARL / HARL

Official implementation of HARL algorithms based on PyTorch.

521 stars 64 forks source link

Cannot reproduce the results for MPE #50

Closed BolunDai0216 closed 1 month ago

BolunDai0216 commented 2 months ago

Hi,

I am trying to reproduce the results for the MPE environments as shown in the image using the default config files.

However, the reward curves for the speaker listener (continuous) environment is very different. Is it possible to share the experiment setup for those curves?

Thanks in advance!

Ivan-Zhong commented 2 months ago

Hi, the experiment setups are detailed in README. To reproduce the results, I recommend using the tuned configs for those experiments, instead of the default configs. Hope it helps. :)