SBX v0.18.0: Bug fix for SAC, optimize log of ent coeff to be consistent with SB3

What's Changed

• Optimize the log of the entropy coeff instead of the entropy coeff by @jamesheald in #56

New Contributors

• @jamesheald made their first contribution in #56

Full Changelog: v0.17.0...v0.18.0

SBX v0.17.0: CNN support for DQN

What's Changed

• Fix warning and remove DroQ class in favor of SAC config by @araffin in #47
• Add CNN support for DQN by @araffin in #49

Full Changelog: v0.15.0...v0.17.0

SBX v0.15.0: Hotfix for offpolicy algorithms, the pseudo random key was not updated

Note

No performance difference should be expected (See report in #46), this bug was introduced in v0.11.0.

What's Changed

• Support for setting the target entropy by @jan1854 in #43
• Hotfix - Return the new updated key in function _train by @theovincent in #46

New Contributors

• @theovincent made their first contribution in #46

Full Changelog: v0.13.0...v0.15.0

SBX v0.13.0: Added CrossQ algorithm and support for custom activations

Warning

Using DroQ class directly is deprecated and will be removed in SBX v0.14.0.
Please use SAC/TQC/CrossQ directly instead with the DroQ configuration, see https://github.com/araffin/sbx?tab=readme-ov-file#note-about-droq

To upgrade:

pip install sbx-rl --upgrade

CrossQ: https://openreview.net/forum?id=PczQtTsTIX (SAC with batch norm and no target network)

What's Changed

• Fix for new tensorflow probability version by @araffin in #39
• Allow to pass custom activation function in policy_kwargs by @paolodelia99 in #41
• Add CrossQ by @araffin, @danielpalen and @jan1854 in #28

New Contributors

• @paolodelia99 made their first contribution in #41
• @danielpalen made their first contribution in #28

Full Changelog: v0.12.0...v0.13.0

SBX v0.12.0: Added support for MultiDiscrete and MultiBinary action spaces to PPO

What's Changed

• Support for MultiDiscrete and MultiBinary action spaces in PPO by @jan1854 in #30

Full Changelog: v0.11.0...v0.12.0

SBX v0.11.0: Added support for large values for gradient_steps to SAC, TD3, and TQC

What's Changed

• Added support for large values for gradient_steps to SAC, TD3, and TQC by @jan1854 in #21

New Contributors

• @jan1854 made their first contribution in #21

Full Changelog: v0.10.0...v0.11.0

SBX v0.10.0: Fix `train()` signature and update type hints

What's Changed

• Fix train signature and update type hints by @araffin in #24

Full Changelog: v0.9.1...v0.10.0

SBX v0.9.1: Fix replay buffer device at load time

What's Changed

• Fix replay buffer device at load time by @araffin in #20

This issue was introduced with SB3 v2.2.1.

Full Changelog: v0.9.0...v0.9.1

SBX v0.9.0: Add flatten layer

What's Changed

• Add flatten layer and update dependencies by @araffin in #18

Full Changelog: v0.8.0...v0.9.0

SBX v0.8.0: Added DDPG and TD3

What's Changed

• Add DDPG and TD3 by @araffin in #16

Full Changelog: v0.7.0...v0.8.0