SB3-Contrib v2.0.0: Gymnasium Support

Warning
Stable-Baselines3 (SB3) v2.0 will be the last one supporting python 3.7 (end of life in June 2023).
We highly recommended you to upgrade to Python >= 3.8.

SB3 Contrib (more algorithms): https://github.com/Stable-Baselines-Team/stable-baselines3-contrib
RL Zoo3 (training framework): https://github.com/DLR-RM/rl-baselines3-zoo
Stable-Baselines Jax (SBX): https://github.com/araffin/sbx

To upgrade:

pip install stable_baselines3 sb3_contrib rl_zoo3 --upgrade

or simply (rl zoo depends on SB3 and SB3 contrib):

pip install rl_zoo3 --upgrade

Breaking Changes

• Switched to Gymnasium as primary backend, Gym 0.21 and 0.26 are still supported via the shimmy package (@carlosluis, @arjun-kg, @tlpss)
• Upgraded to Stable-Baselines3 >= 2.0.0

Bug fixes

• Fixed QRDQN update interval for multi envs

Others

• Fixed sb3_contrib/tqc/*.py type hints
• Fixed sb3_contrib/trpo/*.py type hints
• Fixed sb3_contrib/common/envs/invalid_actions_env.py type hints

Full Changelog: v1.8.0...v2.0.0

SB3-Contrib v1.8.0

Warning
Stable-Baselines3 (SB3) v1.8.0 will be the last one to use Gym as a backend.
Starting with v2.0.0, Gymnasium will be the default backend (though SB3 will have compatibility layers for Gym envs).
You can find a migration guide here.
If you want to try the SB3 v2.0 alpha version, you can take a look at PR #1327.

RL Zoo3 (training framework): https://github.com/DLR-RM/rl-baselines3-zoo

To upgrade:

pip install stable_baselines3 sb3_contrib rl_zoo3 --upgrade

or simply (rl zoo depends on SB3 and SB3 contrib):

pip install rl_zoo3 --upgrade

Breaking Changes:

• Removed shared layers in mlp_extractor (@AlexPasqua)
• Upgraded to Stable-Baselines3 >= 1.8.0

New Features:

• Added stats_window_size argument to control smoothing in rollout logging (@jonasreiher)

Bug Fixes:

Deprecations:

Others:

• Moved to pyproject.toml
• Added github issue forms
• Fixed Atari Roms download in CI
• Fixed sb3_contrib/qrdqn/*.py type hints
• Switched from flake8 to ruff

Documentation:

• Added warning about potential crashes caused by check_env in the MaskablePPO docs (@AlexPasqua)

SB3-Contrib v1.7.0 : Bug fixes for PPO LSTM and quality of life improvements

Warning
Shared layers in MLP policy (mlp_extractor) are now deprecated for PPO, A2C and TRPO.
This feature will be removed in SB3 v1.8.0 and the behavior of net_arch=[64, 64]
will create separate networks with the same architecture, to be consistent with the off-policy algorithms.

Note
TRPO models saved with SB3 < 1.7.0 will show a warning about
missing keys in the state dict when loaded with SB3 >= 1.7.0.
To suppress the warning, simply save the model again.
You can find more info in issue # 1233

Breaking Changes:

• Removed deprecated create_eval_env, eval_env, eval_log_path, n_eval_episodes and eval_freq parameters,
  please use an EvalCallback instead
• Removed deprecated sde_net_arch parameter
• Upgraded to Stable-Baselines3 >= 1.7.0

New Features:

• Introduced mypy type checking
• Added support for Python 3.10
• Added with_bias parameter to ARSPolicy
• Added option to have non-shared features extractor between actor and critic in on-policy algorithms (@AlexPasqua)
• Features extractors now properly support unnormalized image-like observations (3D tensor)
  when passing normalize_images=False

Bug Fixes:

• Fixed a bug in RecurrentPPO where the lstm states where incorrectly reshaped for n_lstm_layers > 1 (thanks @kolbytn)
• Fixed RuntimeError: rnn: hx is not contiguous while predicting terminal values for RecurrentPPO when n_lstm_layers > 1

Deprecations:

• You should now explicitely pass a features_extractor parameter when calling extract_features()
• Deprecated shared layers in MlpExtractor (@AlexPasqua)

Others:

• Fixed flake8 config
• Fixed sb3_contrib/common/utils.py type hint
• Fixed sb3_contrib/common/recurrent/type_aliases.py type hint
• Fixed sb3_contrib/ars/policies.py type hint
• Exposed modules in __init__.py with __all__ attribute (@ZikangXiong)
• Removed ignores on Flake8 F401 (@ZikangXiong)
• Upgraded GitHub CI/setup-python to v4 and checkout to v3
• Set tensors construction directly on the device
• Standardized the use of from gym import spaces

SB3-Contrib v1.6.2: Progress bar

Breaking Changes:

• Upgraded to Stable-Baselines3 >= 1.6.2

New Features:

• Added progress_bar argument in the learn() method, displayed using TQDM and rich packages

Deprecations:

• Deprecate parameters eval_env, eval_freq and create_eval_env

Others:

• Fixed the return type of .load() methods so that they now use TypeVar

SB3-Contrib v1.6.1: Bug fix release

Breaking Changes:

• Fixed the issue that predict does not always return action as np.ndarray (@qgallouedec)
• Upgraded to Stable-Baselines3 >= 1.6.1

Bug Fixes:

• Fixed the issue of wrongly passing policy arguments when using CnnLstmPolicy or MultiInputLstmPolicy with RecurrentPPO (@mlodel)
• Fixed division by zero error when computing FPS when a small number of time has elapsed in operating systems with low-precision timers.
• Fixed calling child callbacks in MaskableEvalCallback (@CppMaster)
• Fixed missing verbose parameter passing in the MaskableEvalCallback constructor (@BurakDmb)
• Fixed the issue that when updating the target network in QRDQN, TQC, the running_mean and running_var properties of batch norm layers are not updated (@honglu2875)

Others:

• Changed the default buffer device from "cpu" to "auto"

sb3-contrib v1.6.0: RecurrentPPO (aka PPO LSTM) and better defaults for learning from pixels with offpolicy algos

Breaking changes:

• Upgraded to Stable-Baselines3 >= 1.6.0
• Changed the way policy "aliases" are handled ("MlpPolicy", "CnnPolicy", ...), removing the former
  register_policy helper, policy_base parameter and using policy_aliases static attributes instead (@Gregwar)
• Renamed rollout/exploration rate key to rollout/exploration_rate for QRDQN (to be consistent with SB3 DQN)
• Upgraded to python 3.7+ syntax using pyupgrade
• SB3 now requires PyTorch >= 1.11
• Changed the default network architecture when using CnnPolicy or MultiInputPolicy with TQC,
  share_features_extractor is now set to False by default and the net_arch=[256, 256] (instead of net_arch=[] that was before)

New Features

• Added RecurrentPPO (aka PPO LSTM)

Bug Fixes:

• Fixed a bug in RecurrentPPO when calculating the masked loss functions (@rnederstigt)
• Fixed a bug in TRPO where kl divergence was not implemented for MultiDiscrete space

sb3-contrib v1.5.0: Bug fixes and newer gym version

Breaking Changes:

• Switched minimum Gym version to 0.21.0.
• Upgraded to Stable-Baselines3 >= 1.5.0

New Features:

• Allow PPO to turn of advantage normalization (see PR #61) @vwxyzjn

Bug Fixes:

• Removed explict calls to forward() method as per pytorch guidelines

sb3-contrib v1.4.0: Trust Region Policy Optimization (TRPO) and Augmented Random Search (ARS) algorithms

Breaking Changes:

• Dropped python 3.6 support
• Upgraded to Stable-Baselines3 >= 1.4.0
• MaskablePPO was updated to match latest SB3 PPO version (timeout handling and new method for the policy object)

New Features:

• Added TRPO (@cyprienc)
• Added experimental support to train off-policy algorithms with multiple envs (note: HerReplayBuffer currently not supported)
• Added Augmented Random Search (ARS) (@sgillen)

Others:

• Improve test coverage for MaskablePPO

sb3-contrib v1.3.0 : PPO with invalid action masking

WARNING: This version will be the last one supporting Python 3.6 (end of life in Dec 2021).
We highly recommended you to upgrade to Python >= 3.7.

Breaking Changes:

• Removed sde_net_arch
• Upgraded to Stable-Baselines3 >= 1.3.0

New Features:

• Added MaskablePPO algorithm (@kronion)
• MaskablePPO Dictionary Observation support (@glmcdona)

sb3-contrib v1.2.0 : Train/Eval mode support

Breaking Changes:

• Upgraded to Stable-Baselines3 >= 1.2.0

Bug Fixes:

• QR-DQN and TQC updated so that their policies are switched between train and eval mode at the correct time (@ayeright)

Others:

• Fixed type annotation
• Added python 3.9 to CI