Summary:
Pull Request resolved: https://github.com/facebookresearch/ReAgent/pull/298
A continuous actor takes in user + N candidate features for PVM
Reviewed By: czxttkl
Differential Revision: D22428244
fbshipit-source-id: e9f8dde9ce16489ee1d0b52cd3b15751077d32f9
Summary: This is the first step to use a distributed data reader: build a torchscript batch preprocessor.
Reviewed By: kaiwenw
Differential Revision: D22036324
fbshipit-source-id: d45980e9cb2863579a1b10d7cfd8c3f2cb610ca5
Summary:
Pull Request resolved: https://github.com/facebookresearch/ReAgent/pull/264
- Move PDQN from fblearner folder to OSS. Fill in ModelManager for it.
- Add Gym tests for maxq and sarsa
- Clean up ParametricBatchPreprocessor to use transforms
- Delete RawTrainingBatch
- Delete open_ai_gym_tests :)
Reviewed By: kittipatv
Differential Revision: D21481434
fbshipit-source-id: b895d3f20d72fb0aa22163c5220654c8b5bfd21b
Summary:
- Removed manual squeezing in seq2slate code path; ensure that the 0th dim of tensor corresponds to batch size
- Use transforms in Seq2Slate batch preprocessor
Reviewed By: czxttkl
Differential Revision: D21224914
fbshipit-source-id: 9935d1641d0f9109e998775385cac821bc8c5a06
Summary: Make it easier to compose transformation
Reviewed By: badrinarayan
Differential Revision: D21198737
fbshipit-source-id: 158b956bc7148e27aeb36108257bccde7502e67f