v0.2.0

What's Changed

• PlaNet implementation by @luisenp in #86
• Changed signature of reset and sample methods to use a model state dictionary by @luisenp in #129
• Feature pybullet by @dtch1997 in #135
• Support pybullet-based Gym Environments by @gauravmm in #87
• [Bug-fix] Made truncated_normal_ completely in-place by @jan1854 in #141
• Change SAC library to use the implementation written by pranz24 by @luisenp in #142
• Removed use of torch.clamp() from icem for compatibility with torch 1.7. by @luisenp in #148
• Removed tensorboard dependencies from pytorch_sac. by @luisenp in #154
• changed path handling to be OS agnostic by @MarcoKe in #153
• Minor doc fix: add missing "cd" statement after cloning repo by @odelalleau in #156
• Fix Hydra warning by sticking to OmegaConf 2.0.x by @odelalleau in #157
• Updated pre-commit configuration and fixed some mypy errors. by @luisenp in #163
• PID code and Update Readme by @natolambert in #165
• Add requirement that importlib_metadata < 5. by @luisenp in #170
• Include dict comprehension by @boneyag in #168
• fix epoch/episode iteration bug in mbpo.py by @matthiaskiller in #160
• Fix format errors by @luisenp in #175
• Fix format and ci by @Rohan138 in #176
• Fix centering, scaling and clamping population in ICEMOptimizer.optimize by @marbaga in #174
• move x to device for gaussian mlp tests by @raghavauppuluri13 in #179
• Gymnasium migration by @Rohan138 in #177

New Contributors

• @dtch1997 made their first contribution in #135
• @gauravmm made their first contribution in #87
• @MarcoKe made their first contribution in #153
• @odelalleau made their first contribution in #156
• @natolambert made their first contribution in #165
• @boneyag made their first contribution in #168
• @matthiaskiller made their first contribution in #160
• @marbaga made their first contribution in #174
• @raghavauppuluri13 made their first contribution in #179

Full Changelog: v0.1.4...v0.2.0

v0.1.5

• Fixes important bug in v0.1.4 that was causing PETS to break.
• Model.reset() and Model.sample() signature has changed. They no longer receive
  TransitionBatch objects, and they both return a dictionary of strings to tensors
  representing a model state that should be passed to sample() to simulate
  transitions. This dictionary can contain things like previous actions, predicted
  observation, latent states, beliefs, and any other such quantity that the model
  need to maintain to simulate trajectories when using ModelEnv.
• Ensemble class and sub-classes are assumed to operate on 1-D models.

v0.1.4

This version adds two new optimizers for CEM:

• Improved CEM as described here.
• MPPI as used in PDDM.
• Changed config structure so that action optimizer is passed as another config file.
• Added a new iterator for sequences that returns a fixed number of random batches in every loop.

v0.1.3

This version changes the Model API so that loss, eval_score and update methods return a metadata dictionary that can be used for logging. It also adds the option to use double precision for normalization.

v0.1.2

This version adds a training browser to the diagnostics folder, and a new buffer iterator for batches of sequences of transitions.

First PyPi version

Merge pull request #82 from facebookresearch/pypi

Refactored to upload to pypi

Initial release

v0.1.0

Initial commit