Expand description
Rollout samplers for r2l on-policy algorithms.
DirectSampler lets workers write directly to trajectory buffers, while
StagedSampler receives transitions from workers and can transform them
before writing them. Both support single-threaded and multi-threaded
environment workers.
Structs§
- Direct
Sampler - Rollout sampler whose workers write directly to output buffers.
- Direct
Sampler Core - Mutable direct-sampler state exposed to
DirectSamplerHookimplementations. - Staged
Sampler - Observation-normalizing rollout sampler controlled by a hook.
- Staged
Sampler Core - Mutable staged-sampler state exposed to hook implementations.
Enums§
- Rollout
Mode - Bound used for one rollout collection request per environment.
- Sampler
Execution Mode - Execution strategy used by the sampler.
- Sampler
Hook Result - Instruction returned by a
DirectSamplerHookduring rollout collection.
Traits§
- Direct
Sampler Hook - Hook that controls the sequence of collection bounds for a raw sampler.
- Staged
Sampler Hook - Hook that controls the sequence of collection bounds for a staged sampler.