Mava
Mava copied to clipboard
Use `TrainState` in evaluator
As discussed in #994 we should now be able to use the MetricsWrapper
and the TrainState
in the evaluator instead of manually recording the episode return and creating a special EvalState
.