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

Cast learning_rate to float lambda for pickle safety when doing model.load #1901

Merged
merged 5 commits into from
Apr 22, 2024
Merged

Cast learning_rate to float lambda for pickle safety when doing model.load #1901

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
dfc0666
create failing test for unpickle error
markscsmith Apr 18, 2024
5ab2fcd
Fix learning_rate argument causing failure in weights_only=True if pa…
markscsmith Apr 19, 2024
10eb400
Updated with feedback from araffin on PR#1901
markscsmith Apr 21, 2024
d35ee6a
Update test and version
araffin Apr 22, 2024
c1d8e60
Update changelog and SBX doc
araffin Apr 22, 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
3 changes: 2 additions & 1 deletion docs/misc/changelog.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,6 +48,7 @@ New Features:
Bug Fixes:
^^^^^^^^^^
- Fixed ``monitor_wrapper`` argument that was not passed to the parent class, and dones argument that wasn't passed to ``_update_into_buffer`` (@corentinlger)
- Fixed ``learning_rate`` argument that could cause weights_only=True to fail if passed a function with non-float types (e.g. ``learning_rate=lambda _: np.sin(1.0)``) (@markscsmith)

`SB3-Contrib`_
^^^^^^^^^^^^^^
Expand Down Expand Up @@ -1593,4 +1594,4 @@ And all the contributors:
@anand-bala @hughperkins @sidney-tio @AlexPasqua @dominicgkerr @Akhilez @Rocamonde @tobirohrer @ZikangXiong @ReHoss
@DavyMorgan @luizapozzobon @Bonifatius94 @theSquaredError @harveybellini @DavyMorgan @FieteO @jonasreiher @npit @WeberSamuel @troiganto
@lutogniew @lbergmann1 @lukashass @BertrandDecoster @pseudo-rnd-thoughts @stefanbschneider @kyle-he @PatrickHelm @corentinlger
@marekm4 @stagoverflow @rushitnshah
@marekm4 @stagoverflow @rushitnshah @markscsmith
2 changes: 1 addition & 1 deletion stable_baselines3/common/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,7 +92,7 @@ def get_schedule_fn(value_schedule: Union[Schedule, float]) -> Schedule:
value_schedule = constant_fn(float(value_schedule))
else:
assert callable(value_schedule)
return value_schedule
return lambda _: float(value_schedule(_))
Copy link
Member

@araffin araffin Apr 19, 2024 •
edited
Loading

Choose a reason for hiding this comment

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

maybe a better solution is to do a call to value_schedule(1.0) and check that the return type is a float (and output a useful error message if not).

Copy link
Member

Choose a reason for hiding this comment

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

or, what you do is fine but I would explicitly name the parameter progress_remaining and add a comment of why

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Hm... I see the value in both. Let me noodle a bit and I'll see if I can sort it out during my lunch. Thanks araffin!



def get_linear_fn(start: float, end: float, end_fraction: float) -> Schedule:
Expand Down
8 changes: 8 additions & 0 deletions tests/test_save_load.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -739,6 +739,14 @@ def test_load_invalid_object(tmp_path):
assert len(record) == 0


def test_learning_rate_float_for_unpickle(tmp_path):
path = str(tmp_path / "ppo_pendulum.zip")
PPO("MlpPolicy", "Pendulum-v1", learning_rate=lambda _: np.sin(1.0)).save(path)
with warnings.catch_warnings(record=True) as record:
PPO.load(path)
assert len(record) == 0


def test_dqn_target_update_interval(tmp_path):
# `target_update_interval` should not change when reloading the model. See GH Issue #1373.
env = make_vec_env(env_id="CartPole-v1", n_envs=2)
Expand Down