Skip to content

Fix advantage calculation off-by-one error (#363) #445

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jacobarrio wants to merge 1 commit into
base: 3.0
Choose a base branch
from
Open

Fix advantage calculation off-by-one error (#363) #445

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 10 additions & 10 deletions pufferlib/pufferl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,19 +87,19 @@ def __init__(self, config, vecenv, policy, logger=None):
)

device = config['device']
self.observations = torch.zeros(segments, horizon, *obs_space.shape,
self.observations = torch.zeros(segments, horizon + 1, *obs_space.shape,
dtype=pufferlib.pytorch.numpy_to_torch_dtype_dict[obs_space.dtype],
pin_memory=device == 'cuda' and config['cpu_offload'],
device='cpu' if config['cpu_offload'] else device)
self.actions = torch.zeros(segments, horizon, *atn_space.shape, device=device,
self.actions = torch.zeros(segments, horizon + 1, *atn_space.shape, device=device,
dtype=pufferlib.pytorch.numpy_to_torch_dtype_dict[atn_space.dtype])
self.values = torch.zeros(segments, horizon, device=device)
self.logprobs = torch.zeros(segments, horizon, device=device)
self.rewards = torch.zeros(segments, horizon, device=device)
self.terminals = torch.zeros(segments, horizon, device=device)
self.truncations = torch.zeros(segments, horizon, device=device)
self.ratio = torch.ones(segments, horizon, device=device)
self.importance = torch.ones(segments, horizon, device=device)
self.values = torch.zeros(segments, horizon + 1, device=device)
self.logprobs = torch.zeros(segments, horizon + 1, device=device)
self.rewards = torch.zeros(segments, horizon + 1, device=device)
self.terminals = torch.zeros(segments, horizon + 1, device=device)
self.truncations = torch.zeros(segments, horizon + 1, device=device)
self.ratio = torch.ones(segments, horizon + 1, device=device)
self.importance = torch.ones(segments, horizon + 1, device=device)
self.ep_lengths = torch.zeros(total_agents, device=device, dtype=torch.int32)
self.ep_indices = torch.arange(total_agents, device=device, dtype=torch.int32)
self.free_idx = total_agents
Expand Down Expand Up @@ -282,7 +282,7 @@ def evaluate(self):

# Note: We are not yet handling masks in this version
self.ep_lengths[env_id] += 1
if l+1 >= config['bptt_horizon']:
if l+1 > config['bptt_horizon']:
num_full = env_id.stop - env_id.start
self.ep_indices[env_id] = self.free_idx + torch.arange(num_full, device=config['device']).int()
self.ep_lengths[env_id] = 0
Expand Down