instadeepai / Mava

🦁 A research-friendly codebase for fast experimentation of multi-agent reinforcement learning in JAX
Apache License 2.0
737 stars 90 forks source link

[FEATURE] Register for Jumanji environments #975

Closed WiemKhlifi closed 10 months ago

WiemKhlifi commented 10 months ago

Feature

Implement a registry function/class for Jumanji environments to avoid creating numerous configuration files for each scenario (e.g. for RWARE and LBF). In this function, we check the existence of the given scenario and create the corresponding attributes to initialise the environment class.

Proposal

For the selected environment-scenario pair, we first check the existence of the scenario in an initialized data structure (dictionary or list) of valid scenarios containing all possible correct combinations. This data structure will be automatically initialised for the given environment, making it easier to extract attribute names and values to create the env object.

Testing

Test functionality when creating the selected environment-scenario pair with most scenarios.

Definition of done

This function can be considered complete after being tested on most of the valid scenarios that can be initialised for RWARE and LBF at present.

Mandatory checklist before making a PR

Links / references / screenshots

WiemKhlifi commented 10 months ago

Check the Jumanji register PR for further information.