Summary:
Pull Request resolved: https://github.com/facebookresearch/ReAgent/pull/299
This diff accomplishes several items:
1. Remove PageHandler and consolidate all training functions into one function, using polymorphism to handle model-specific logic
2. Make BatchRunner the sole place where FB vs. OSS context is decided (by choosing FbBatchRunner or OssBatchRunner)
3. Transform ModelManager into a stateless provider.
4. With the exception of model manager, remove all duplicate classes by creating oss & internal versions and using polymorphism, or moving out of workflow/* entirely
5. Replace signals-and-slots API with interfaces
6. Create a DataFetcher class, unifying the APIs to query data on OSS and FB.
Reviewed By: kaiwenw
Differential Revision: D22702504
fbshipit-source-id: 3eb8e93144ca12ac650a4fafc875e29d8ade89e3
Summary: this class not used anywhere
Reviewed By: kittipatv
Differential Revision: D22105436
fbshipit-source-id: 4948d2813e7be1369958ddbb6bd8a7ca212ad645
Summary:
- Convert rest of models to use new registry, deprecating RLTrainer methods like set_optimizer, get_optimizer
- make TrainerParameters with make_config_class
Reviewed By: kittipatv
Differential Revision: D21759010
fbshipit-source-id: b18e72f149670371c2784edc19bb9da0337b0308
Summary:
Updated CB RASP tutorial to use new OSS workflow.
Introduce FileSystemPublisher, a key-value database with tinydb, an OSS version of predictor. This way, after training the path to torchscript can be uploaded, and during evaluation it can be read.
Pull Request resolved: https://github.com/facebookresearch/ReAgent/pull/270
Reviewed By: MisterTea
Differential Revision: D21689872
fbshipit-source-id: e8d510fe895dcb33d41464e026af842e31366f87
Summary:
- Delete SingleQValue and AllActionQValues; returns tensor directly
- CategoricalDQN is now taking a ModelBase which returns 3-D tensor (batch_size, num_actions, num_atoms)
- FullyConnectedDQN's constructor has one more argument, num_atoms, which default to None. When it's not None, returns 3-D tensor like above
- DuelingQNetwork takes 3 sub networks: shared_network, value_network, advantage_network. The parametric version is split out to ParametricDuelingQNetwork for better readability.
- Introduces a subclass of Sequential conforming to ModelBase API. `input_prototype()` returns the value from the first module.
Reviewed By: czxttkl
Differential Revision: D21336534
fbshipit-source-id: 629c688b0de4754effb30cd61f4a6da3cca021ab
Summary:
No need to build custom docker image anymore.
- Use sdkman to install compilers for Spark app
- Use miniconda to install build deps for RASP
Pull Request resolved: https://github.com/facebookresearch/ReAgent/pull/225
Reviewed By: MisterTea
Differential Revision: D20974491
Pulled By: kittipatv
fbshipit-source-id: 680ab3f60c6c63aea5854f99242243b5f1d348bc
Summary:
Fix the usage example such that it obtains a reasonable policy.
Pull Request resolved: https://github.com/facebookresearch/ReAgent/pull/196
Test Plan:
Imported from GitHub, without a `Test Plan:` line.
Existing tests pass.
Manually ran `run_end_to_end_test.sh` and it passed.
Reviewed By: kittipatv
Differential Revision: D19078228
Pulled By: czxttkl
fbshipit-source-id: d50f4fd258583635d16b087f324cd75d18931200
Summary: We no longer use Thrift for configuration
Reviewed By: czxttkl
Differential Revision: D18242481
fbshipit-source-id: bfdfa1e7baa21be92bb51a7103d0f379dafee2b2
Summary: Missing a new line causing the block not to render
Reviewed By: czxttkl
Differential Revision: D17999728
fbshipit-source-id: 20af5affbbe8b0bd54c7261d424cf9b033bb5e3d
Summary: Add a page for CI instructions
Reviewed By: czxttkl
Differential Revision: D17983742
fbshipit-source-id: 31479a349310a55844cea7ea97629fcd96d5b1b6
Summary:
Update several things so that sphinx compiled successfully:
conf.py (add onnx in autodoc_mock_imports and add Horizon root to sys.path)
ml/rl/caffe_utils.py (update __getattr__ such that sphnix wouldn't throw exception)
ml/rl/tensorboardX.py (update __getattr__ such that sphnix wouldn't throw exception)
Pull Request resolved: https://github.com/facebookresearch/Horizon/pull/146
Test Plan: Wait for all integration tests and unit tests pass
Reviewed By: kittipatv
Differential Revision: D16243745
Pulled By: czxttkl
fbshipit-source-id: e3b8af37ec520bde44a4a59379341bceaa981070
Zhengxing Chen
Jason Gauci
Kaiwen Wang
Stanislau Hlebik
Kai Wen Wang
Kittipat Virochsiri
Zhengxing Chen
Brett Koonce
Edoardo Conti
Jason Gauci
baek-jinoo