[Feature] SAC compatibility with composite distributions. #2447

albertbou92 · 2024-09-22T22:21:41Z

Description

This PR adapts the SAC objective to be compatible with composite distributions.

DiscreteSAC seems a little bit more tricky, as it requires computing the log probabilities of all possible actions. In the case of composite distributions this might mean the probabilities of all combinations of actions.

Motivation and Context

Why is this change required? What problem does it solve?
If it fixes an open issue, please link to the issue here.
You can use the syntax close #15213 if this solves the issue #15213

I have raised an issue to propose this change (required for new features and bug fixes)

Types of changes

What types of changes does your code introduce? Remove all that do not apply:

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds core functionality)
Breaking change (fix or feature that would cause existing functionality to change)
Documentation (update in the documentation)
Example (update in the folder of examples)

Checklist

Go over all the following points, and put an x in all the boxes that apply.
If you are unsure about any of these, don't hesitate to ask. We are here to help!

I have read the CONTRIBUTION guide (required)
My change requires a change to the documentation.
I have updated the tests accordingly (required for a bug fix or a new feature).
I have updated the documentation accordingly.

pytorch-bot · 2024-09-22T22:21:45Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/2447

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 12 New Failures, 7 Unrelated Failures

As of commit 53ba371 with merge base f411f93 ():

NEW FAILURES - The following jobs have failed:

Continuous Benchmark (PR) / CPU Pytest benchmark (gh)
/home/runner/work/_temp/c86bd8ef-c8cd-4daf-a072-483ebbf48a64.sh: line 5: 2083 Bus error (core dumped) $RUN_BENCHMARK /tmp/tmp.OoQCxxnhhD
Continuous Benchmark (PR) / GPU Pytest benchmark (gh)
Process completed with exit code 135.
Continuous Benchmark / CPU Pytest benchmark (gh)
##[error]Process completed with exit code 135.
Habitat Tests on Linux / tests (3.9, 12.1) / linux-job (gh)
RuntimeError: Command docker exec -t 02d4594e1efe06710b3b83683844631d7e110cae6af0e0ae607209f29f61b862 /exec failed with exit code 134
Unit-tests on Linux / tests-cpu (3.10) / linux-job (gh)
RuntimeError: Command docker exec -t 6521d5a192ddfe0a2c6c9aed3e2fd0c19914f988a8bdb063ea9fb5cf1a2b0017 /exec failed with exit code 1
Unit-tests on Linux / tests-cpu (3.11) / linux-job (gh)
RuntimeError: Command docker exec -t 2c91baf0aa1bc8342cc9dcfa660689dc0ade546141375dcf91243a35371bdd13 /exec failed with exit code 1
Unit-tests on Linux / tests-cpu (3.12) / linux-job (gh)
RuntimeError: Command docker exec -t 73c878c590147c7b42e356ac87a9b5ad21074741242bd80c277856f6b92b7f1d /exec failed with exit code 1
Unit-tests on Linux / tests-cpu (3.9) / linux-job (gh)
RuntimeError: Command docker exec -t 3662c0816bbf1203cd39c205aae1ebf03134266e35da40571171fc4e0c2436bf /exec failed with exit code 1
Unit-tests on Linux / tests-cpu-oldget (3.12) / linux-job (gh)
RuntimeError: Command docker exec -t 8b83084a536f43f53aae0b29562bdb03dbee86f0b3813da0f0da93221230dc4c /exec failed with exit code 1
Unit-tests on Linux / tests-gpu (3.11, 12.1) / linux-job (gh)
RuntimeError: Command docker exec -t 5240947f34dc66839e0485c8ce2ec3a818d9bb7b3704bc03d942b71d27a02bc2 /exec failed with exit code 1
Unit-tests on Linux / tests-optdeps (3.10, 12.1) / linux-job (gh)
RuntimeError: Command docker exec -t b831325816747b76e5b9632a812c88028ba80762ffee448f01e532cec61fe84b /exec failed with exit code 1
Unit-tests on Linux / tests-stable-gpu (3.10, 11.8) / linux-job (gh)
RuntimeError: Command docker exec -t 59691caffc5d201a93acfa70c539ad26e631131965cbb76bdca77619ee9055de /exec failed with exit code 1

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

Build Windows Wheels / pytorch/rl / upload / wheel-py3_9-cpu (gh) (trunk failure)
##[error]Unable to find any artifacts for the associated workflow
Build Windows Wheels / pytorch/rl / upload / wheel-py3_9-cuda11_8 (gh) (trunk failure)
##[error]Unable to find any artifacts for the associated workflow
Build Windows Wheels / pytorch/rl / upload / wheel-py3_9-cuda12_1 (gh) (trunk failure)
##[error]Unable to find any artifacts for the associated workflow
Build Windows Wheels / pytorch/rl / upload / wheel-py3_9-cuda12_4 (gh) (trunk failure)
##[error]Unable to find any artifacts for the associated workflow
Continuous Benchmark / GPU Pytest benchmark (gh) (trunk failure)
Fatal Python error: Fatal Python error: Fatal Python error: Fatal Python error: Bus errorBus errorBus error
Unit-tests on Linux / tests-olddeps (3.8, 11.6) / linux-job (gh) (trunk failure)
test/test_tensordictmodules.py::TestGRUModule::test_gru_vmap_complex_model
Unit-tests on Windows / unittests-cpu / windows-job (gh) (trunk failure)
RuntimeError: Compiler: cl is not found.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

albertbou92 · 2024-09-22T22:23:50Z

torchrl/objectives/sac.py

@@ -251,7 +264,7 @@ class _AcceptedKeys:
            state_action_value (NestedKey): The input tensordict key where the
                state action value is expected.  Defaults to ``"state_action_value"``.
            log_prob (NestedKey): The input tensordict key where the log probability is expected.
-                Defaults to ``"_log_prob"``.
+                Defaults to ``"sample_log_prob"``.


I renamed this key to make it work with the new API. It did not seem to break anything, but I imagine there was a reason for the naming.

albertbou92 · 2024-09-22T22:32:41Z

torchrl/objectives/sac.py

@@ -450,9 +463,7 @@ def target_entropy(self):
            else:
                action_container_shape = action_spec.shape
            target_entropy = -float(
-                action_spec[self.tensor_keys.action]


This too I had to change. Shouldn't action spec already have the correct shape?

facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Sep 22, 2024

albertbou92 commented Sep 22, 2024

View reviewed changes

albertbou92 added 8 commits October 10, 2024 14:56

tests sac composite dist

99e845c

test fixes

1bfaf6c

test fixes

c936df8

test fixes

cdb91ec

test fixes

9e5eb09

test fixes

c8c9cc8

rm copy

cf98dbe

fix

53ba371

vmoens force-pushed the offline_compositedis branch from 886705d to 53ba371 Compare October 10, 2024 13:56

vmoens added the enhancement New feature or request label Oct 10, 2024

vmoens merged commit ec04c35 into pytorch:main Oct 11, 2024
60 of 79 checks passed

vmoens deleted the offline_compositedis branch October 11, 2024 15:18

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Feature] SAC compatibility with composite distributions. #2447

[Feature] SAC compatibility with composite distributions. #2447

albertbou92 commented Sep 22, 2024 •

edited

Loading

pytorch-bot bot commented Sep 22, 2024 •

edited

Loading

albertbou92 Sep 22, 2024

albertbou92 Sep 22, 2024

[Feature] SAC compatibility with composite distributions. #2447

[Feature] SAC compatibility with composite distributions. #2447

Conversation

albertbou92 commented Sep 22, 2024 • edited Loading

Description

Motivation and Context

Types of changes

Checklist

pytorch-bot bot commented Sep 22, 2024 • edited Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/2447

❌ 12 New Failures, 7 Unrelated Failures

albertbou92 Sep 22, 2024

Choose a reason for hiding this comment

albertbou92 Sep 22, 2024

Choose a reason for hiding this comment

albertbou92 commented Sep 22, 2024 •

edited

Loading

pytorch-bot bot commented Sep 22, 2024 •

edited

Loading