dodox/metastable-baselines
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fixed Bug: Wrong dimensions for action_loss

Browse Source
This commit is contained in:
Dominik Moritz Roth 2022-09-03 11:44:01 +02:00
parent 4bb772a251
commit 0aeea4e2e5
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
metastable_baselines/ppo/ppo.py
View File

@ -330,12 +330,14 @@ class PPO(GaussianRolloutCollectorAuxclass, OnPolicyAlgorithm):
trust_region_losses.append(trust_region_loss.item()) trust_region_losses.append(trust_region_loss.item())
# 'Principle of least action' # 'Principle of least action'
action_loss = th.square(actions) action_loss = th.mean(th.square(actions))
action_losses.append(action_loss) action_losses.append(action_loss)
policy_loss = surrogate_loss + self.ent_coef * entropy_loss + \ policy_loss = surrogate_loss + self.ent_coef * entropy_loss + \
trust_region_loss + self.action_coef * action_loss trust_region_loss + self.action_coef * action_loss
import pdb
pdb.set_trace()
pg_losses.append(policy_loss.item()) pg_losses.append(policy_loss.item())
loss = policy_loss + self.vf_coef * value_loss loss = policy_loss + self.vf_coef * value_loss