steffen/marl-factory-grid
1
0
Fork 0
mirror of https://github.com/illiumst/marl-factory-grid.git synced 2025-07-05 17:11:35 +02:00
Code Issues Packages Projects Releases Wiki Activity

Resolved some warnings and style issues

Browse Source
This commit is contained in:
Steffen Illium
2023-11-10 09:29:54 +01:00
parent a9462a8b6f
commit 6711a0976b
64 changed files with 331 additions and 361 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
marl_factory_grid/algorithms/marl/mappo.py
View File

@ -36,7 +36,7 @@ class LoopMAPPO(LoopSNAC):
rewards_ = torch.stack(rewards_, dim=1)
return rewards_
def mappo(self, batch, network, gamma, entropy_coef, vf_coef, clip_range, **kwargs):
def mappo(self, batch, network, gamma, entropy_coef, vf_coef, clip_range, **__):
out = network(batch[nms.OBSERVATION], batch[nms.ACTION], batch[nms.HIDDEN_ACTOR], batch[nms.HIDDEN_CRITIC])
logits = out[nms.LOGITS][:, :-1] # last one only needed for v_{t+1}
@ -45,7 +45,7 @@ class LoopMAPPO(LoopSNAC):
# monte carlo returns
mc_returns = self.monte_carlo_returns(batch[nms.REWARD], batch[nms.DONE], gamma)
mc_returns = (mc_returns - mc_returns.mean()) / (mc_returns.std() + 1e-8) #todo: norm across agent ok?
mc_returns = (mc_returns - mc_returns.mean()) / (mc_returns.std() + 1e-8) # todo: norm across agent ok?
advantages = mc_returns - out[nms.CRITIC][:, :-1]
# policy loss