Changelog [skip-ci]
This commit is contained in:
parent
6aa33b1bfe
commit
e2e8a699ea
@ -22,6 +22,7 @@ Launchers for parallelization currently in alpha state. #1074
|
||||
- Base class for collectors: `BaseCollector` #1122
|
||||
- Collectors can now explicitly specify whether to use the policy in training or evaluation mode. #1122
|
||||
- New util context managers `in_eval_mode` and `in_train_mode` for torch modules. #1122
|
||||
- `reset` of `Collectors` now returns `obs` and `info`. #1122
|
||||
|
||||
### Internal Improvements
|
||||
- `Collector`s rely less on state, the few stateful things are stored explicitly instead of through a `.data` attribute. #1063
|
||||
|
Loading…
x
Reference in New Issue
Block a user