Pytorch Version for Mean Field Multi-Agent Reinforcement Learning

Pytorch implementation of MF-Q and MF-AC in the paper Mean Field Multi-Agent Reinforcement Learning.

The original code can be found in mlii/mfrl.

Please uncomment the following two lines of code in base.py if the algorithm occasionally fails to converge.

  #distribution = torch.distributions.Categorical(predict) 
  #actions = distribution.sample().detach().cpu().numpy()

Example