huangfu/InteractionImitation
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Rename predict to forward

Browse Source 
This is done to be consistent with stable baselines interface. predict is then automatically defined. This is necessary to use stable baselines' evaluate_policy method
This commit is contained in:
Johannes Fischer
2021-10-28 09:42:27 +02:00
parent 71f69c43ed
commit 3857716cec
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/gail/options.py
View File

@@ -36,7 +36,7 @@ class OptionsEnv(gym.Wrapper):
self.episode_start = True self.episode_start = True
self.m = available_actions(self.env, self.options) self.m = available_actions(self.env, self.options)
self.ch, self.value, self.log_prob = generator.policy.predict({ self.ch, self.value, self.log_prob = generator.policy.forward({
'obs': torch.tensor(self.s).unsqueeze(0).to(generator.policy.device), 'obs': torch.tensor(self.s).unsqueeze(0).to(generator.policy.device),
'mask': torch.tensor(self.m).unsqueeze(0).to(generator.policy.device), 'mask': torch.tensor(self.m).unsqueeze(0).to(generator.policy.device),
}) })

2
src/policies/options.py
View File

@@ -22,7 +22,7 @@ class OptionsCnnPolicy(stable_baselines3.common.policies.ActorCriticCnnPolicy):
values = self.value_net(latent_vf) values = self.value_net(latent_vf)
return values, distribution.distribution return values, distribution.distribution
def predict(self, obs, eps=1e-6): def forward(self, obs, eps=1e-6):
""" """
Will mask invalid states before making action selections Will mask invalid states before making action selections
Args: Args: