V0.3.2 - Githubissues

ai4co / rl4co

A PyTorch library for all things Reinforcement Learning (RL) for Combinatorial Optimization (CO)

https://rl4.co

MIT License

386 stars 71 forks source link

V0.3.2 #116

Closed fedebotu closed 6 months ago

fedebotu commented 7 months ago

Bug Fixes

SMTWTP environment with batch size initialization
RL4CO base env: default batch size to None; improvements

QOL

can pass "None" for no baseline in REINFORCE

Performance

Improve by changing torch.no_grad -> [torch.inference_mode](https://pytorch.org/docs/stable/generated/torch.inference_mode.html)
Faster testing

Versioning

Prepare v0.3.2 (will also include new Beam Search and decoding API)