instadeepai / Mava

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

chore: minor evaluator clean up #1015

Closed sash-a closed 7 months ago

sash-a commented 7 months ago

What?

Minor clean up to the evaluator.

Removed the creation of trained params and eval keys from the eval setup method as it's not really the responsibility of that method, it should just make the evaluation methods.

Also created the unreplicate_batch_dim method to easily unreplicate our update_batch_size dim.

Finally removed the unecessary vmap inside the evaluator as flax will automatically vmap for us