Miscellaneous BugFixes and Features - Githubissues

ai4co / rl4co

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

https://rl4.co

MIT License

381 stars 70 forks source link

Miscellaneous BugFixes and Features #186

Closed fedebotu closed 3 months ago

fedebotu commented 3 months ago

Description

Feats

A2C: fully modular optimizers for actor and critic
Linear attention (experimental) support

BugFixes

Optimize all parameters by default in RL4COLitModule (to allow for e.g. A2C and out-of-policy training)
Fix inverted sign of critic loss

Configs

Standardize defaults for routing
Add A2C