rlberry.manager.MultipleManagers

class rlberry.manager.MultipleManagers(max_workers: int | None = None, parallelization: str = 'process', mp_context='spawn')

Bases: object

Class to fit multiple ExperimentManager instances in parallel with multiple threads.

Parameters:

max_workers: int, default=None

max number of workers (ExperimentManager instances) fitted at the same time.

parallelization: {‘thread’, ‘process’}, default: ‘process’

Whether to parallelize agent training using threads or processes.

mp_context: {‘spawn’, ‘fork’, ‘forkserver’}, default: ‘spawn’.

Context for python multiprocessing module. Warning: If you’re using JAX or PyTorch, it only works with ‘spawn’.

If running code on a notebook or interpreter, use ‘fork’.

Attributes:

managers

Methods

append(experiment_manager)	Append new ExperimentManager instance.
run([save])	Fit ExperimentManager instances in parallel.
save()	Pickle ExperimentManager instances and saves fit statistics in .csv files.

append(experiment_manager)

Append new ExperimentManager instance.

Parameters:

experiment_managerExperimentManager

run(save=True)

Fit ExperimentManager instances in parallel.

Parameters:

save: bool, default: True

If true, save ExperimentManager intances immediately after fitting. ExperimentManager.save() is called.

save()

Pickle ExperimentManager instances and saves fit statistics in .csv files. The output folder is defined in each of the ExperimentManager instances.