v0.2.0
Enhancements
- Add type hints to DQN and its dependencies (#17)
- Pretrained models (#33)
- Optuna integration & example script (2) (#46)
- Save and load a module trained with DataParallel or DistributedDataParallel (#55)
- Replace nn.functional.tanh with torch.tanh (#64)
- Makes common write_header function (#71)
- Eval stats collector (#80)
- Set stats type (#83)
- Add
step_hooks
andoptimizer_step_hooks
to actor-learner (#86) - Adds env_stats to record_tb_stats (#98)
Bug fixes
- calculate scalar q_values to log average_q statistics in categorical dqn algorithms (#32)
- Detach greedy_actions before calling cpu() (#81)
- Fix the hang in
train_agent_async
with Tensorboard (#89)
Examples
- Pretrained models (#33)
- [Fix Examples] function call was duplicated (#36, thanks @gorogoroyasu!)
- Optuna integration & example script (2) (#46)
Testing