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()