evorl.algorithms.erl.erl_td3.erl_origin

Module Contents

Classes

ERLTrainMetric

ERLWorkflow

Original ERL impl.

Functions

get_ec_pop_statistics

API

class evorl.algorithms.erl.erl_td3.erl_origin.ERLTrainMetric[source]

Bases: evorl.algorithms.erl.erl_workflow.ERLTrainMetric

num_updates_per_iter: chex.Array

‘zeros(…)’

time_cost_per_iter: float

0.0

class evorl.algorithms.erl.erl_td3.erl_origin.ERLWorkflow(**kwargs)[source]

Bases: evorl.algorithms.erl.erl_td3.erl_ga.ERLGAWorkflow

Original ERL impl.

Have dynamic training updates per iteration, i.e., #rl_updates = #sampled_timesteps_this_iter. Therefore the step() function cannot be directly jitted.

classmethod enable_jit() → None[source]
classmethod name()[source]
step(state: evorl.types.State) → tuple[evorl.metrics.MetricBase, evorl.types.State][source]

The basic step function for the workflow to update agent.

evorl.algorithms.erl.erl_td3.erl_origin.get_ec_pop_statistics(pop)[source]