Expand description
Experience buffers and replay management (requires training feature).
The storage layer is plain Vec<f32>/Vec<i64> regardless of which
tensor backend the trainer ultimately uses; tensor materialization
happens via to_burn_tensors on the batch types when the trainer
calls into Burn.
Experience buffers and replay management
This module handles storage and sampling of experience for training.
Re-exports§
pub use rollout::compute_advantages_multi_agent;