Skip to content
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

[BugFix] Allow for composite action distributions in PPO/A2C losses #2391

Merged
merged 19 commits into from
Sep 4, 2024
Merged

[BugFix] Allow for composite action distributions in PPO/A2C losses #2391

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
627b673
account for composite distribution
albertbou92 Aug 10, 2024
9b08555
tests
albertbou92 Aug 11, 2024
a668d0c
fix tests ppo
albertbou92 Aug 12, 2024
e65e1a3
fix tests ppo
albertbou92 Aug 12, 2024
fd60ad7
fix tests ppo
albertbou92 Aug 12, 2024
bc00e2d
fix tests ppo
albertbou92 Aug 12, 2024
e3f61f7
fix tests ppo
albertbou92 Aug 12, 2024
4eb413d
fix tests ppo
albertbou92 Aug 12, 2024
560ddad
fix tests ppo
albertbou92 Aug 12, 2024
2865c92
format
albertbou92 Aug 12, 2024
13a092d
a2c tests
albertbou92 Aug 14, 2024
8bc8e03
a2c tests
albertbou92 Aug 14, 2024
07ec262
a2c tests
albertbou92 Aug 14, 2024
a95eb3d
a2c tests
albertbou92 Aug 14, 2024
8f00828
format
albertbou92 Aug 14, 2024
9ce3c58
docstrings and minor fixes
albertbou92 Aug 14, 2024
ad32b90
docstrings and minor fixes
albertbou92 Aug 14, 2024
37b733f
docstrings fix
albertbou92 Aug 14, 2024
69922fa
added required grad for td
albertbou92 Aug 26, 2024
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
Prev Previous commit
Next Next commit
a2c tests
  • Loading branch information
@albertbou92
albertbou92 committed Aug 14, 2024
commit 07ec262ee298f5e2d1eea643e36009a2c33c5ff8
16 changes: 13 additions & 3 deletions test/test_cost.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8860,14 +8860,24 @@ def test_a2c(
functional=functional,
)

def set_requires_grad(tensor, requires_grad):
tensor.requires_grad = requires_grad
return tensor

# Check error is raised when actions require grads
td["action"].requires_grad = True
if composite_action_dist:
td["action"].apply_(lambda x: set_requires_grad(x, True))
else:
td["action"].requires_grad = True
with pytest.raises(
RuntimeError,
match="tensordict stored action require grad.",
match="tensordict stored action requires grad.",
):
_ = loss_fn._log_probs(td)
td["action"].requires_grad = False
if composite_action_dist:
td["action"].apply_(lambda x: set_requires_grad(x, False))
else:
td["action"].requires_grad = False

td = td.exclude(loss_fn.tensor_keys.value_target)
if advantage is not None:
Expand Down
16 changes: 11 additions & 5 deletions torchrl/objectives/a2c.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -384,7 +384,7 @@ def get_entropy_bonus(self, dist: d.Distribution) -> torch.Tensor:
except NotImplementedError:
x = dist.rsample((self.samples_mc_entropy,))
log_prob = dist.log_prob(x)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Previously, was this a bug or did we sum the log-probs automatically?

Copy link
Contributor Author

@albertbou92 albertbou92 Aug 26, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is simply because the log_prob() method for a composite dist will return a TD instead of a Tensor, so we compute the entropy in 2 steps.

This is the old version:

    def get_entropy_bonus(self, dist: d.Distribution) -> torch.Tensor:
        try:
            entropy = dist.entropy()
        except NotImplementedError:
            x = dist.rsample((self.samples_mc_entropy,))
            entropy = -dist.log_prob(x).mean(0)
        return entropy.unsqueeze(-1)

This is the new version. It simply retrieves the log tensor before computing the entropy.

    def get_entropy_bonus(self, dist: d.Distribution) -> torch.Tensor:
        try:
            entropy = dist.entropy()
        except NotImplementedError:
            x = dist.rsample((self.samples_mc_entropy,))
            log_prob = dist.log_prob(x)
            if is_tensor_collection(log_prob):
                log_prob = log_prob.get(self.tensor_keys.sample_log_prob)
            entropy = -log_prob.mean(0)
        return entropy.unsqueeze(-1)

if isinstance(x, log_prob):
if isinstance(log_prob, TensorDict):
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A lazy stack is not a TensorDict but a TensorDict base.
Also ideally we would want this to work with tensorclasses.
The way to go should be to use is_tensor_collection from tensordict lib.

log_prob = log_prob.get(self.tensor_keys.sample_log_prob)
entropy = -log_prob.mean(0)
return entropy.unsqueeze(-1)
Expand All @@ -394,20 +394,26 @@ def _log_probs(
) -> Tuple[torch.Tensor, d.Distribution]:
# current log_prob of actions
action = tensordict.get(self.tensor_keys.action)
if action.requires_grad:
raise RuntimeError(
f"tensordict stored {self.tensor_keys.action} require grad."
)
tensordict_clone = tensordict.select(
*self.actor_network.in_keys, strict=False
).clone()
with self.actor_network_params.to_module(
self.actor_network
) if self.functional else contextlib.nullcontext():
dist = self.actor_network.get_dist(tensordict_clone)

def check_requires_grad(tensor):
if tensor.requires_grad:
albertbou92 marked this conversation as resolved.
Show resolved Hide resolved
raise RuntimeError(
f"tensordict stored {self.tensor_keys.action} requires grad."
)
return tensor

if isinstance(action, torch.Tensor):
check_requires_grad(action)
log_prob = dist.log_prob(action)
else:
action.apply(check_requires_grad)
tensordict = dist.log_prob(tensordict)
log_prob = tensordict.get(self.tensor_keys.sample_log_prob)
log_prob = log_prob.unsqueeze(-1)
Expand Down
14 changes: 8 additions & 6 deletions torchrl/objectives/ppo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -460,24 +460,26 @@ def _log_weight(
) -> Tuple[torch.Tensor, d.Distribution]:
# current log_prob of actions
action = tensordict.get(self.tensor_keys.action)
if action.requires_grad:
raise RuntimeError(
f"tensordict stored {self.tensor_keys.action} requires grad."
)

with self.actor_network_params.to_module(
self.actor_network
) if self.functional else contextlib.nullcontext():
dist = self.actor_network.get_dist(tensordict)
# dist = TransformedDistribution(dist, ExpTransform())

def check_requires_grad(tensor, key=self.tensor_keys.action):
albertbou92 marked this conversation as resolved.
Show resolved Hide resolved
if tensor.requires_grad:
raise RuntimeError(f"tensordict stored {key} requires grad.")
return tensor

prev_log_prob = tensordict.get(self.tensor_keys.sample_log_prob)
if prev_log_prob.requires_grad:
raise RuntimeError("tensordict prev_log_prob requires grad.")
check_requires_grad(prev_log_prob, self.tensor_keys.sample_log_prob)

if isinstance(action, torch.Tensor):
check_requires_grad(action, self.tensor_keys.action)
log_prob = dist.log_prob(action)
else:
action.apply(check_requires_grad)
tensordict = dist.log_prob(tensordict)
log_prob = tensordict.get(self.tensor_keys.sample_log_prob)

Expand Down