Module offline 

Read-only offline dataset buffer for offline RL algorithms.

Unlike [ReplayBuffer], this buffer is loaded once from a static dataset and never modified. It supports:

• Uniform i.i.d. transition sampling (for TD3+BC, IQL, CQL, BC)
• Trajectory subsequence sampling (for Decision Transformer)
• Return-conditioned sampling (for return-conditioned methods)
• Dataset normalization statistics

Designed for D4RL/Minari-scale datasets (1M+ transitions).

Structs

DatasetStats

Statistics about the loaded dataset.

OfflineBatch

A batch of i.i.d. sampled transitions.

OfflineDatasetBuffer

Read-only offline dataset buffer.

TrajectoryBatch

A batch of contiguous trajectory subsequences.