How to implement a OnPolicy RL Strategy?

[raresvanca]

I tried implementing a OnPolicyStrategy for a PPO model by looking at how crates/burn-train/src/learner/rl/off_policy.rs is done, but it seems that I need RLEvent or RLEventProcessorType, that are not exported from the crate. Am I missing something?

[laggui]

Oh you're right. Looks like it's currently limited to pub(crate) as opposed to the other train event types which are publicly re-exported. That means any custom strategy won't be able to leverage all training components like the event processor.

[laggui]

I created an issue from this discussion: #4946