Expand description
Generic implementation of replay buffers for reinforcement learning.
This module provides a flexible and efficient implementation of replay buffers that can handle arbitrary observation and action types. It supports both standard experience replay and prioritized experience replay (PER).
§Key Components
SimpleReplayBuffer: A generic replay buffer implementationGenericTransitionBatch: A generic batch structure for transitionsSimpleStepProcessor: A processor for converting environment steps to transitionsPerConfig: Configuration for prioritized experience replay
§Features
- Generic type support for observations and actions
- Efficient batch processing
- Prioritized experience replay with importance sampling
- Configurable weight normalization
- Step processing for non-vectorized environments
Structs§
- Generic
Transition Batch - A generic structure representing transitions in reinforcement learning.
- IwScheduler
- Scheduler of the exponent of importance weight for PER.
- PerConfig
- Configuration for Prioritized Experience Replay (PER).
- Simple
Replay Buffer - A generic implementation of a replay buffer for reinforcement learning.
- Simple
Replay Buffer Config - Configuration for the replay buffer.
- Simple
Step Processor - A generic implementation of the
StepProcessortrait. - Simple
Step Processor Config - Configuration for the simple step processor.
Enums§
- Weight
Normalizer - Specifies how to normalize the importance weights in a prioritized batch.
Traits§
- Batch
Base - A trait defining basic batch operations.