Nightly release

Merged PRs since last release:

🐛 Fixes

• tgolsson: use input_key in FeatureAgentProxy.input_names (#90)
• tgolsson: allow None in cycle to disable (#91)
• tgolsson: Add missing dataclass decorator on MetaData (#92)
• tgolsson: Improve error handling in memory (#98)
• tgolsson: fix tests (#103)
• spetravic: Missing return statement in AlphaLoss (#111)
• tgolsson: add atomicwrites dependency (#113)
• AliGhadirzadeh: Simona ali/tridays (#112)
• tgolsson: ensure memory exports are readable everywhere (#116)
• spetravic: Hyperbolic tangent instability (#118)
• tgolsson: ensure pending exports are processed per batch (#120)
• tgolsson: work around memory leak in torch.jit.trace (#126)
• tgolsson: detach images (#127)
• jaxs-ribs: Fix simple doc error (#131)
• jaxs-ribs: Add protocol=4 to cloudpickle.dump (#134)
• spetravic: Simona/fix logging (#144)
• tgolsson: update pdm-plugin-torch (#145)
• spetravic: Simona/onnx (#161)
• spetravic: Simona/add conditions (#162)

🧪 Tests

• tgolsson: fix tests (#103)

🧪 Development

• singhblom: Updated a few things missed in 3.8=>3.9 (#102)
• tgolsson: simplify lockfile error (#105)
• tgolsson: Improve CI error messages (#107)
• tgolsson: missing init file (#106)
• tgolsson: add final step to trigger other repo (#108)
• ghost: Ali/mbrl tests (#110)
• tgolsson: Add PDM 2.5.0 to CI and bump minimum version to 2.4.0 (#115)
• tgolsson: Fix branch filter for docs publish (#119)

Uncategorized:

• tgolsson: Add documentation about writing docs (#94)
• tgolsson: refactor LoggingCallback to LoggingMixin (#93)
• ghost: Gym is replaced by gymnasium (#95)
• singhblom: Ignore pypackages (#100)
• tgolsson: add LoggingProxyWrapper for the memory (#96)
• spetravic: Weights&biases logging integration (#104)
• ghost: Ali/mbrl lib (#101)
• AliGhadirzadeh: fixed the input_key error (#114)
• tgolsson: remove unused exitsignal (#124)
• tgolsson: Upgrade Memory Warmup callback to handle exports and early shutdowns (#125)
• jaxs-ribs: Add checkpointer_index constructor argument to Checkpointer (#128)
• tgolsson: fix version to work around urllib breakage (#121)
• jaxs-ribs: Add Storage Subdirectory (#130)
• tgolsson: split callbacks.py into a package (#109)
• spetravic: Add support for custom weight decay optimizer (#129)
• tgolsson: update torch to 1.12 (#133)
• tgolsson: Enable setting device for ONNX exporter (#135)
• jaxs-ribs: Add more logging to ckpt/onnx/memory (#136)
• tgolsson: always flush after writing tb events (#138)
• tgolsson: check for double-init episodes (#123)
• spetravic: Add logging (#140)
• tgolsson: fix docs build? (#141)
• spetravic: Agent proxy that supports multiple input keys (#143)
• spetravic: Add Conv1d encoder and update Conv2d encoder with option to flatten (#142)
• AliGhadirzadeh: updating the checkpointer and checkpoint loader (#146)
• klashenriksson: Fix alpha loss load_state_dict fn signature (#147)
• spetravic: Switch order of checks in log_windowed_scalar (#148)
• spetravic: Update windowed scalar (again) (#149)
• singhblom: Remove _inf_step from tb logs (#151)
• singhblom: Orthogonal init for conv as well (#152)
• spetravic: Change check for bias initialization (#156)
• tgolsson: add PDM 2.6 and 2.7 to CI image (#153)
• klashenriksson: add optional per-parameter logging for loss callbacks (#158)
• tgolsson: setup multi-python image (#160)
• spetravic: Simona/inits (#163)
• klashenriksson: Make MemoryProxyWrappers forward methods to wrapee if they do not exist on wrapper (#164)
• tgolsson: simplify agent proxy setup (#165)
• tgolsson: cherry-pick changes to reward-test-check (#166)
• tgolsson: Upgrade memory serialization workflows (#167)
• tgolsson: add OnnxExporter.add_metadata (#168)
• AliGhadirzadeh: Ali/genrl pr (#159)
• tgolsson: use same line-length as other repositories (#171)
• tgolsson: split state restoration from begin_training (#172)
• tgolsson: run flake8 on CI (#173)
• klashenriksson: Allow temp storage to grow to fit the largest requested data (#174)
• jaxs-ribs: Fix Lunar Lander Test (#177)
• jaxs-ribs: Implement DQN and CartPole Test (#175)
• tgolsson: log memory timers (#179)
• jaxs-ribs: Implement Coverage Based Sampling Strategy (#178)
• jaxs-ribs: Add "latest_checkpoint" path to state_dict (#169)
• jaxs-ribs: Add Documentation for Cartpole Training (#180)
• spetravic: Add schedules (#185)
• spetravic: Rename get_last_val (#187)
• spetravic: Update AlphaLoss to support entropy schedules (#186)
• jaxs-ribs: Fix wrong onnx.load_model arguments (#188)
• tgolsson: fix bug in logging timers (#189)
• tgolsson: add pdm 2.9 and 2.10 to CI image (#190)
• jaxs-ribs: Upgrade to Python 3.10 (#184)
• tgolsson: Upgrade to PDM 2.10 (#191)
• jaxs-ribs: Update Github Workflows for Python3.10 (#192)
• jaxs-ribs: Properly set workflow file (#193)
• klashenriksson: Make LoggingProxyWrapper report inf_step if requested (#194)
• AliGhadirzadeh: Ali/amp (#195)
• klashenriksson: Add a joint memory loader capable of sampling data from multiple memory loaders (#181)
• AliGhadirzadeh: adding JointMemoryLoader to the init (#196)
• klashenriksson: Add NextNElementWrapper virtual storage (#198)

Release v23.0.0

23.0.0 - 2023-03-03

Breaking

• The minimum required Python version is now 3.9 (#87)
• The torch-cpu feature has been renamed to torch as it wasn't limited to CPU-only variants. (#76)

Added

• Enable exporting ONNX policies for Gaussian MLPs in emote.extra.onnx_exporter. This allows you to peridiocally write ONNX files to disk. (#80)
• Add system performance logger in emote.extra.system_logger. This'll log memory and CPU usage to Tensorboard. (#81)
• Add memory warmup waiter in emote.memory.memory to ensure the memory has enough data before starting to sample. This avoids the collector having to block the training loop when training async. (#78)

Changed

• Our PDM plugin for torch management has been split off into a separate repository and published to PYPI. (#88)
• Switch to PDM 2.3 as default version for testing (#62)
• The input key used for Feature Agent Proxies can now be customized (#79)

Release refs/tags/v22.0.0

Merged PRs since last release:

🚀 Features

• singhblom: Msb/classes in memory (#3)
• singhblom: Changed cycle length to be callback specific (#5)
• singhblom: Made checkpointing work and tested (#6)
• riley-mld: Make train on lunar lander (#14)
• jackharmer: n-step TD, gpu training, episodic rewards, high LR nan fix. (#30)
• jackharmer: Fixes SAC algorithm logic. Simplifies distribution code. Fixes terminal adaptor reshaping. (#38)
• riley-mld: Add learning rate schedule (#39)
• jackharmer: video logging (#42)
• tgolsson: Add plugin to manage torch versions (#58)

🧪 Development

• tgolsson: Add docs publishing pipeline and switch back to poetry (#28)
• tgolsson: Run black and isort on CI (#31)
• tgolsson: Sanity-checks (#33)
• tgolsson: Don't install cuda-version of torch on CI (#35)
• tgolsson: Check line-endings in CI and remove all CRLF (#34)
• tgolsson: Use a custom base image for CI runs (#36)
• tgolsson: bump poetry in CI (#45)
• tgolsson: Testing bandit (#43)
• tgolsson: switch to PDM (#49)
• tgolsson: fix doc publish (#50)
• tgolsson: Add package and test steps (#44)
• tgolsson: Add pre-release with auto-generated changelog and GHP docs (#51)

Uncategorized:

• singhblom: Removed unnecessary SAC Network class (#1)
• singhblom: Made open source readme changes (#4)
• singhblom: Rename to Emote (#7)
• singhblom: Moved to Conda (#8)
• singhblom: New nicer readme (#9)
• repi: Minor README visual tweaks (#10)
• singhblom: Removed all mentions of Hive (#11)
• singhblom: Ignore the runs dir where tb logs (#15)
• tgolsson: delete conda env and pypi file (#32)
• jackharmer: Jack/vision (#48)
• tgolsson: ADR: Release flow (#55)
• tgolsson: Add ADRs to deployed documentation (#60)
• tgolsson: Create release flow on-tag (#57)
• tgolsson: prepare changelog for release (#61)