ugr-sail / sinergym

Gym environment for building simulation and control using reinforcement learning
https://ugr-sail.github.io/sinergym/
MIT License
131 stars 35 forks source link

[Enhancement]: WandB migration from callbacks to wrapper #426

Closed AlejandroCN7 closed 2 months ago

AlejandroCN7 commented 3 months ago

Improvement 🔧

To address compatibility issues with Sinergym environments, it is proposed to migrate all real-time training registration functionality to a native Sinergym Wrapper, rather than implementing it by inheriting from the classes provided by Stable Baselines 3.

Environment interaction log in real-time should be as general as possible for better compatibility with the broader family of algorithms. Using the original callback with a WandB adapter only to log specific algorithm metrics during training.

This migration would simplify all functionality as there would be no need to adapt to the particularities and issues of SB3.

Checklist

:pencil: Please, don't forget to include more labels besides enhancement if it is necessary.