Expand description
Sum-tree backed prioritized experience replay.
Provides O(log N) prefix-sum sampling and priority updates for Prioritized Experience Replay (Schaul et al., 2015).
Structs§
- LAPConfig
- Loss-Adjusted Prioritization (LAP) configuration.
- Prioritized
Replay Buffer - Prioritized experience replay buffer backed by a sum-tree.
- Prioritized
Sampled Batch - A sampled batch with importance-sampling weights.
- SumTree
- Binary sum-tree for O(log N) prefix-sum queries.
Functions§
- compute_
lap_ priorities - Compute LAP priorities from TD errors and losses.
- compute_
td_ priorities - Convenience: compute priorities from TD errors only (standard PER).