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] Computation of log prob in composite distribution for TensorDict without batch size #1065

Merged
merged 2 commits into from
Nov 1, 2024
Merged

[BugFix] Computation of log prob in composite distribution for TensorDict without batch size #1065

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
761f359
Fix bug on the forward pass with CompositeDistribution
Oct 30, 2024
b247384
Apply suggestions from code review
priba Oct 31, 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
10 changes: 5 additions & 5 deletions tensordict/nn/probabilistic.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -392,17 +392,17 @@ def forward(
if _requires_sample:
out_tensors = self._dist_sample(dist, interaction_type=interaction_type())
if isinstance(out_tensors, TensorDictBase):
tensordict_out.update(out_tensors)
if self.return_log_prob:
kwargs = {}
if isinstance(dist, CompositeDistribution):
kwargs = {"aggregate_probabilities": False}
log_prob = dist.log_prob(tensordict_out, **kwargs)
if log_prob is not tensordict_out:
log_prob = dist.log_prob(out_tensors, **kwargs)
if log_prob is not out_tensors:
# Composite dists return the tensordict_out directly when aggrgate_prob is False
tensordict_out.set(self.log_prob_key, log_prob)
out_tensors.set(self.log_prob_key, log_prob)
else:
tensordict_out.rename_key_(dist.log_prob_key, self.log_prob_key)
out_tensors.rename_key_(dist.log_prob_key, self.log_prob_key)
tensordict_out.update(out_tensors)
else:
if isinstance(out_tensors, Tensor):
out_tensors = (out_tensors,)
Expand Down
134 changes: 132 additions & 2 deletions test/test_nn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2331,7 +2331,7 @@ def test_prob_module(self, interaction, return_log_prob, map_names):
},
}
},
[3],
batch_size=(3,),
)
in_keys = ["params"]
out_keys = ["cont", ("nested", "cont")]
Expand Down Expand Up @@ -2389,6 +2389,82 @@ def test_prob_module(self, interaction, return_log_prob, map_names):
+ sample_clone.get(key_logprob1).sum(-1),
)

@pytest.mark.parametrize(
"interaction", [InteractionType.MODE, InteractionType.MEAN]
)
@pytest.mark.parametrize("map_names", [True, False])
def test_prob_module_nested(self, interaction, map_names):
params = TensorDict(
{
"agents": TensorDict(
{
"params": {
"cont": {
"loc": torch.randn(3, 4, requires_grad=True),
"scale": torch.rand(3, 4, requires_grad=True),
},
("nested", "cont"): {
"loc": torch.randn(3, 4, requires_grad=True),
"scale": torch.rand(3, 4, requires_grad=True),
},
}
},
batch_size=3,
),
"done": torch.ones(1),
}
)
in_keys = [("agents", "params")]
out_keys = ["cont", ("nested", "cont")]
distribution_map = {
"cont": distributions.Normal,
("nested", "cont"): distributions.Normal,
}
distribution_kwargs = {
"distribution_map": distribution_map,
"log_prob_key": ("agents", "sample_log_prob"),
}
if map_names:
distribution_kwargs.update(
{
"name_map": {
"cont": ("sample", "agents", "cont"),
("nested", "cont"): ("sample", "agents", "nested", "cont"),
}
}
)
out_keys = list(distribution_kwargs["name_map"].values())
module = ProbabilisticTensorDictModule(
in_keys=in_keys,
out_keys=None,
distribution_class=CompositeDistribution,
distribution_kwargs=distribution_kwargs,
default_interaction_type=interaction,
return_log_prob=True,
log_prob_key=("agents", "sample_log_prob"),
)
# loosely checks that the log-prob keys have been added
assert module.out_keys[-2:] != out_keys

sample = module(params)
key_logprob0 = (
("sample", "agents", "cont_log_prob") if map_names else "cont_log_prob"
)
key_logprob1 = (
("sample", "agents", "nested", "cont_log_prob")
if map_names
else ("nested", "cont_log_prob")
)
assert key_logprob0 in sample
assert key_logprob1 in sample
assert all(key in sample for key in module.out_keys)

lp = sample.get(module.log_prob_key)
torch.testing.assert_close(
lp,
sample.get(key_logprob0).sum(-1) + sample.get(key_logprob1).sum(-1),
)

@pytest.mark.parametrize(
"interaction", [InteractionType.MODE, InteractionType.MEAN]
)
Expand All @@ -2407,7 +2483,7 @@ def test_prob_module_seq(self, interaction, return_log_prob):
},
}
},
[3],
batch_size=(3,),
)
in_keys = ["params"]
out_keys = ["cont", ("nested", "cont")]
Expand Down Expand Up @@ -2446,6 +2522,60 @@ def test_prob_module_seq(self, interaction, return_log_prob):
+ sample_clone.get(("nested", "cont_log_prob")).sum(-1),
)

@pytest.mark.parametrize(
"interaction", [InteractionType.MODE, InteractionType.MEAN]
)
def test_prob_module_seq_nested(self, interaction):
params = TensorDict(
{
"agents": TensorDict(
{
"params": {
"cont": {
"loc": torch.randn(3, 4, requires_grad=True),
"scale": torch.rand(3, 4, requires_grad=True),
},
("nested", "cont"): {
"loc": torch.randn(3, 4, requires_grad=True),
"scale": torch.rand(3, 4, requires_grad=True),
},
}
},
batch_size=3,
),
"done": torch.ones(1),
}
)
in_keys = [("agents", "params")]
out_keys = ["cont", ("nested", "cont")]
distribution_map = {
"cont": distributions.Normal,
("nested", "cont"): distributions.Normal,
}
log_prob_key = ("agents", "sample_log_prob")
backbone = TensorDictModule(lambda: None, in_keys=[], out_keys=[])
module = ProbabilisticTensorDictSequential(
backbone,
ProbabilisticTensorDictModule(
in_keys=in_keys,
out_keys=out_keys,
distribution_class=CompositeDistribution,
distribution_kwargs={"distribution_map": distribution_map},
default_interaction_type=interaction,
return_log_prob=True,
log_prob_key=log_prob_key,
),
)
sample = module(params)
assert "cont_log_prob" in sample.keys()
assert ("nested", "cont_log_prob") in sample.keys(True)
lp = sample[log_prob_key]
torch.testing.assert_close(
lp,
sample.get("cont_log_prob").sum(-1)
+ sample.get(("nested", "cont_log_prob")).sum(-1),
)


class TestAddStateIndependentNormalScale:
def test_add_scale_basic(self, num_outputs=4):
Expand Down
Loading