* Update Gymnasium to v1.0.0a1
* Comment out `gymnasium.wrappers.monitor` (todo update to VideoRecord)
* Fix ruff warnings
* Register Atari envs
* Update `getattr` to `Env.get_wrapper_attr`
* Reorder imports
* Fix `seed` order
* Fix collecting `max_steps`
* Copy and paste video recorder to prevent the need to rewrite the vec vide recorder wrapper
* Use `typing.List` rather than list
* Fix env attribute forwarding
* Separate out env attribute collection from its utilisation
* Update for Gymnasium alpha 2
* Remove assert for OrderedDict
* Update setup.py
* Add type: ignore
* Test with Gymnasium main
* Remove `gymnasium.logger.debug/info`
* Fix github CI yaml
* Run gym 0.29.1 on python 3.10
* Update lower bounds
* Integrate video recorder
* Remove ordered dict
* Update changelog
---------
Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
* Update documentation
Added comment to PPO documentation that CPU should primarily be used unless using CNN as well as sample code. Added warning to user for both PPO and A2C that CPU should be used if the user is running GPU without using a CNN, reference Issue #1245.
* Add warning to base class and add test
---------
Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
* Add np.ndarray as a recognized type for TB histograms.
Torch histograms allow th.Tensor, np.ndarray, and caffe2 formatted strings. This commits expands the TensorBoardOutputFormat's capabilities to log the two former types.
* Update changelog to reflect bug fix
* fix: try/catch for if either np or torch aren't at the required versions. See https://github.com/DLR-RM/stable-baselines3/pull/1635 for more details
* fix: Add comment describing the test for when add_histogram should not have been called
* Cleanup
---------
Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
* Add support for pre and post linear modules in `create_mlp`
* Disable mypy for python 3.8
* Reformat toml file
* Update docstring
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
* Add some comments
---------
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
* Fixing #1791
* Update test and version
* Add test for callback after eval
* Fix mypy error
* Remove tqdm warnings
---------
Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
* Fix loading a model with net_arch=None
* Remove redundant get
* Dummy commit
* Add to contributors
* Update test and version
---------
Co-authored-by: Antonin Raffin <antonin.raffin@ensta.org>
* create failing test for unpickle error
* Fix learning_rate argument causing failure in weights_only=True if passed a function with non-float types
* Updated with feedback from araffin on PR#1901
* Update test and version
* Update changelog and SBX doc
---------
Co-authored-by: Antonin Raffin <antonin.raffin@ensta.org>
* Add success rate in monitor for on policy algorithms
* Update changelog
* make commit-checks refactoring
* Assert buffers are not none in _dump_logs
* Automatic refactoring of the type hinting
* Add success_rate logging test for on policy algorithms
* Update changelog
* Reformat
* Fix tests and update changelog
---------
Co-authored-by: Antonin Raffin <antonin.raffin@ensta.org>
* Add rollout_buffer_class and rollout_buffer_kwargs parameters to OnPolicyAlgorithm
* Add rollout_buffer_class and rollout_buffer_kwargs to PPO.
* Add rollout_buffer_class and rollout_buffer_kwargs to A2C.
* Make use of the rollout buffer kwargs.
* Update version
* Add test and update doc
---------
Co-authored-by: Antonin Raffin <antonin.raffin@dlr.de>
* Update signatures, and test with options
* Update changelog and black formatting
* Finish implementation (fixes, doc, tests)
* Use deepcopy to avoid side effects (modif by reference)
* Fix for mypy
---------
Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
* fix: Follow PEP8 guidelines and evaluate falsy to truth with `not` rather than `is False`.
https://docs.python.org/2/library/stdtypes.html#truth-value-testing
* chore: Update changelog inline with intent of changes in PR #1707
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
* fix: Change `is False` to `not` as per PEP8
* chore: Remove superfluous comment about `is False`
* test: One On- and one Off-Policy algorithm (A2C and SAC respectively), with settings to speed up testing
* Update changelog
* chore: Remove EvalCallback as it's not actually required
* Update changelog.rst
* Rm duplicated "others" section in changelog.rst
---------
Co-authored-by: Quentin Gallouédec <45557362+qgallouedec@users.noreply.github.com>
Co-authored-by: Antonin Raffin <antonin.raffin@ensta.org>
* prevents squash_output if not use_sde, see #1592
* update changelog
* add unscaling of actions taken during training
* add test regarding squashing and unquashing
* avoids try-except block
* format Gymnasium code with black
* makes mypy pass
* makes pytype pass
* sort imports
* makes error message in assert statement clearer
Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
* improves code commenting
* replaces full env with wrapper
* Cleanup code
* Reformat
---------
Co-authored-by: PatrickHelm <patrick.helm@gmx.net>
Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
Co-authored-by: Antonin Raffin <antonin.raffin@dlr.de>
* Fix type hints in `common/utils.py`
* Fix `VecTranspose` type annotations
* Fix types for callbacks
* Update changelog
* Fix video recorder type hints
* Fix save utils type hints
* Allow BytesIO
* Improve error message
* Make logger and training env properties
* Clarify which open_path fn is called
* Fix bug in env_checker.py bounds warning message
* Fix bug where Gym Environment Checker does not output the correct warning message when dealing with observation spaces that have different upper and different lower bounds
* Update test_env_checker.py with more comprehensive tests
* Make naming consistent
* Update version
* Catch all invalid indices at once
---------
Co-authored-by: gabo_tor <gabriel0torre@gmail.com>
* Added test cases where off policy algorithms fail with float64 actionspace
* casting observations and actions to `np.float32` to unify behaviour between `ReplayBuffer` and `RolloutBuffer`. Fixing issue #1145
* reformatted using black
* making test more restrictive by checking models action is float64
* added changelog entry
* undo cast of observations as `preprocessing.preprocess_obs()` casts them to float32 anyways.
* - Casting to float32 only, if action.dtype is float64
- Added cast to `DictReplayBuffer` as well
* Added tests for multiple variations of continuous action types and observation spaces
* applied reformatting by `make commit-checks`
* Added typing and comment referring to description in merge request
* Apply linter for single element slice
* Rename helper and refactor tests
* Update changelog and docstring
---------
Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
* Update setup.py to v0.29.0
* Remove invalid test
* Loosen version and update changelog
---------
Co-authored-by: Antonin Raffin <antonin.raffin@ensta.org>
* Fix env checker single-step-env edge case
Before this change, env checker failed to `reset()` the tested
environment before calling `step()` when checking for `Inf` / `NaN`.
This could cause environments which happened to have only one `step()`
available before the episode was terminated to fail.
This is now fixed.
* Code review fixes#1
As suggested by Antonin Raffin <antonin.raffin@ensta.org>.
