instadeepai / Mava

šŸ¦ A research-friendly codebase for fast experimentation of multi-agent reinforcement learning in JAX
Apache License 2.0
709 stars 83 forks source link

[FEATURE]: Reduce config files #1033

Open WiemKhlifi opened 7 months ago

WiemKhlifi commented 7 months ago

Please describe the purpose of the feature. Is it related to a problem?

Integrate the systems' logger default into a single yaml file and move the common training controlling flags from systems files into the arch yaml files (e.g. seed and total_timesteps). This helps avoid setting things in each system in case we want to run a sweep for example and avoid adding a logger_default for each system.

Describe the solution you'd like

Loggers merged in one file and setting the common training flags outside each system šŸ˜„

How do we know when the implementation of this feature is complete?

Checklist:

Additional context

Check this PR #987 for some inspiration (An example was started in this PR but for now, we keep the systems' default separate) šŸ™Œ

šŸŒŸ Additional request: Merge system defaults for the same system type (such PPOs) and find the best way to select the system config without changing the default.