Crate r2l_core 

Core traits and data types shared by the r2l workspace.

r2l-core is the contracts crate. It defines the small set of interfaces that environments, samplers, policies, agents, learners, and tensor backends agree on. Backend-specific implementations live in crates such as r2l-burn and r2l-candle; concrete algorithms and builders live outside this crate as well.

Most downstream code should start with the prelude:

use r2l_core::prelude::*;

The main extension points are:

• Env and EnvBuilder for environment integrations.
• R2lTensor for tensor types used by environments and learning code.
• Actor, Policy, ValueFunction, and Learner for model and optimizer components.
• TrajectoryBuffer and TrajectoryView for rollout storage.
• Agent, Sampler, and OnPolicyAlgorithm for on-policy training loops.

Modules

buffers

Rollout transition and trajectory storage.

env

Environment traits and space descriptions.

error

Error types

models

Actor, policy, value-function, and learner traits.

on_policy

Shared interfaces for on-policy training loops. Shared contracts for on-policy algorithms.

prelude

Common imports for implementing environments, policies, agents, samplers, and learners.

rng

Reproducible random-number utilities.

running_mean

Online mean and variance estimators.

tensor

Backend-neutral tensor interfaces and adapters.

Macros

break_on_hook_result

Breaks out of the surrounding loop when a hook requests HookResult::Break.

return_on_hook_result

Returns Ok(()) from the surrounding function when a hook requests HookResult::Break.

Structs

ActorWrapper

Enums

HookResult

Control-flow result returned by training hooks.