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

adding stablebaselines, adding safe loading for nonCuda cluster

Browse Source
This commit is contained in:
Arec Jamgochian
2022-02-24 23:18:02 -08:00
parent 5a5d8a7aff
commit 7feea74eb8
2 changed files with 11 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
requirements.txt
View File

@@ -8,3 +8,4 @@ ray[tune]
hyperopt
psutil
fire
stable_baselines3

15
src/eval_main.py
View File

@@ -37,21 +37,22 @@ def load_policy(method:str,
Returns:
policy (Optional[BaseAlgorithm]): the policy to evaluate
"""
ml = torch.device('cpu') if not torch.cuda.is_available() else None
if method == 'idm':
policy = IDMRulePolicy(env, **policy_kwargs)
elif method == 'bc':
policy = SetPolicy(env.action_space.shape[-1])
policy.load_state_dict(torch.load(policy_file))
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
elif method == 'gail':
policy = SetPolicy(env.action_space.shape[-1])
policy(torch.zeros(env.observation_space.shape))
policy = ReparamPolicy(policy)
policy.load_state_dict(torch.load(policy_file))
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
elif method == 'gail-ppo':
policy = SetPolicy(env.action_space.shape[-1])
policy.load_state_dict(torch.load(policy_file))
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
elif method == 'rail':
raise NotImplementedError
@@ -59,11 +60,11 @@ def load_policy(method:str,
policy = SetDiscretePolicy(env.action_space.n)
policy(torch.zeros(env.observation_space.shape))
policy = ReparamPolicy(policy)
policy.load_state_dict(torch.load(policy_file))
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
elif method == 'ogail-ppo':
policy = SetDiscretePolicy(env.action_space.n)
policy.load_state_dict(torch.load(policy_file))
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
elif method == 'sgail':
policy = SetMaskedDiscretePolicy(env.action_space.n)
@@ -72,11 +73,11 @@ def load_policy(method:str,
torch.zeros(env.observation_space['safe_actions'].shape)
)
policy = ReparamSafePolicy(policy)
policy.load_state_dict(torch.load(policy_file))
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
elif method == 'sgail-ppo':
policy = SetMaskedDiscretePolicy(env.action_space.n)
policy.load_state_dict(torch.load(policy_file))
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
else:
raise NotImplementedError