Expand description
Experience replay buffers. Experience replay buffers for off-policy RL algorithms.
Provides three implementations:
crate::buffer::UniformReplayBuffer— uniform random sampling (DQN, SAC, TD3)crate::buffer::PrioritizedReplayBuffer— proportional PER with segment tree sampling (PER-DQN, PER-SAC)crate::buffer::NStepBuffer— n-step return accumulation with discount rollout
Re-exports§
pub use n_step::NStepBuffer;pub use n_step::NStepTransition;pub use prioritized::PrioritizedReplayBuffer;pub use prioritized::PrioritySample;pub use replay::Transition;pub use replay::UniformReplayBuffer;
Modules§
- n_step
- N-step Return Buffer
- prioritized
- Prioritized Experience Replay (PER)
- replay
- Uniform Replay Buffer