An easydict with pytree support.

Evaluate the agent in the environments. Attributes: env: Vectorized environment w/o autoreset. action_fn: The agent action functi

State of the agent. Attributes: params: The network parameters of the agent. obs_preprocessor_state: The state of the observation

Data container for trajectory data.

Continuous space in R^n. Attributes: low: The lower bounds of the box. high: The upper bounds of the box.

A general State class. An alias of PyTreeDict. This class is specfically used for `Workflow` state.

Specification for an exponential schedule for HyperParam.

ReplayBuffer with uniform sampling. Data are added and sampled in 1d-like structure. Attributes: capacity: the maximum capacity of t

Evaluate and return eval metrics and episodic trajectory. EpisodeCollector is similar as `Evaluator`, but it also returns the trajectories. When

Evaluation metrics for RLWorkflow. Attributes: episode_returns: The return array of evaluation episodes. episode_lengths: The len

Discrete space in {0, 1, ..., n-1}. Attributes: n: The number of discrete values.

State of the environment. Include all the data needed to represent the state of the environment. Attributes: env_state: The internal

Vectorize env.

An agent that takes uniform random actions.

MLP module.

State for the prioritized replay buffer. Attributes: priority: Priority values for each entry in the buffer. max_priority: Curren

Vectorized one-episode wrapper for evaluation.

Maintains episode step count and sets done at episode end. When call step() after the env is done, stop simulation and directly return previo

Save the structure of the parameters. Provide methods to convert between the original tree-like parameter and the flatten parameter vector.

Container for structural spaces. Attributes: spaces: a pytree of spaces.

Maintains episode step count and sets done at episode end. This is the same as brax's EpisodeWrapper, and add some new fields in transition.info.

Brax-style AutoReset: no randomness in reset. This wrapper reuses the state in the return of `env.reset()`. When the episodes have short length o

Contains training state for the learner.

Training metrics for RLWorkflow. Attributes: train_episode_return: The return of the training episode. loss: The loss value of th

Vectorize env and Autoreset.

EnvPool style AutoReset. When the episode ends, an additional reset step is performed. See EnvPool: https://envpool.readthedocs.io/en/latest/

Adapter class to convert EvoX algorithms to EvoRL optimizers.

Log file recorder.

OpenAI ES.

Scale the reward by a factor. Usage: - Use EpisodeWrapper(RewardScaleWrapper(env)) to get the scaled `info.episode_return`. - Use RewardS

Workflow metrics for RLWorkflow. Attributes: sampled_timesteps: The total number of sampled timesteps from environments. iteratio

Augmented Random Search. Paper: [Simple random search of static linear policies is competitive for reinforcement learning](https://proceedings.ne

Repeat action for a number of steps. :::{note} This wrapper only accumulates `state.reward` and `state.info.ori_reward`. It is safe to use `A

Deterministic Agent for continuous action space in [-1, 1].

Distributed information for multi-devices training.

Contains training state for the learner.

Training metrics for ECWorkflow. Attributes: objectives: The objective values for current step. ec_metrics: The extra metrics of

Streamlined episode collector for observation only.

State of the OpenES.

Q Module for continuous action space.

Full state of running statistics computation.

Sep Cross-Entropy Method.

Convert dense reward to sparse reward. The dense rewards become: 0, 0, ..., sum(rewards), 0, 0, ..., sum(rewards)

Recorder for Weights & Biases.

Contains training state for the learner.

State of the ARS.

Convert continuous action space from [-1, 1] to [low, high].

Linear Actor Head.

Adapter for Brax environments.

Mutli-objective evaluator for Brax environments. Attributes: metric_names: The names of the metrics to evaluate, default is ("reward", "e

CMA-ES. Paper: [Completely Derandomized Self-Adaptation in Evolution Strategies](http://www.cmap.polytechnique.fr/~nikolaus.hansen/cmaartic.pdf)

A simple CNN backbone used for the agent and the critic.

Container for multiple recorders.

Lineat Critic Head.

The Agnet for DDPG.

Contains training state for the learner.

A dummy checkpoint manager that does nothing.

Workflow metrics for ECWorkflow. Attributes: best_objective: The best objective value found so far. sampled_episodes: The total n

Adapter for EnvPool to support EnvPool environments. This env already is a vectorized environment and has experimental supports. It is not recomm

Adapter for Gymnasium to support Gymnasium environments. This env already is a vectorized environment and has experimental supports. It is not re

Adapter for Gymnax environments.

Contains training state for the learner.

Adapter for Jumanji environments.

ReplayBuffer with Loss-Adjusted Prioritization from TD7 paper. LAP is a variation of Prioritized Experience Replay (PER) that uses proportio