`evorl.algorithms.meta.pbt_openes.param_openes`¶

Module Contents¶

Classes¶

`OpenES`
`ParamOpenESWorkflow`

API¶

class evorl.algorithms.meta.pbt_openes.param_openes.OpenES[source]¶

Bases: evorl.ec.optimizers.EvoOptimizer

ask(state: evorl.ec.optimizers.ECState) → tuple[chex.ArrayTree, evorl.ec.optimizers.ECState][source]¶: Generate new candidate solutions.

fitness_shaping_fn: collections.abc.Callable[[chex.Array], chex.Array]¶: ‘pytree_field(…)’

init(mean: evorl.types.Params, key: chex.PRNGKey) → evorl.ec.optimizers.ECState[source]¶

lr: float¶: None

mirror_sampling: bool¶: True

noise_std: float¶: None

optimizer: optax.GradientTransformation¶: ‘pytree_field(…)’

pop_size: int¶: None

tell(state: evorl.ec.optimizers.ECState, fitnesses: chex.Array) → tuple[evorl.types.PyTreeDict, evorl.ec.optimizers.openes.OpenESState][source]¶

weight_decay: float | None¶: None

class evorl.algorithms.meta.pbt_openes.param_openes.ParamOpenESWorkflow(*, env: evorl.envs.Env, agent: evorl.agent.Agent, ec_optimizer: evorl.ec.optimizers.EvoOptimizer, ec_evaluator: evorl.evaluators.Evaluator, evaluator: evorl.evaluators.Evaluator, agent_state_vmap_axes: evorl.agent.AgentStateAxis = 0, config: omegaconf.DictConfig)[source]¶

Bases: evorl.algorithms.ec.so.openes.OpenESWorkflow

classmethod name()[source]¶