Experiment Runner (Core)#

alts.core.experiment_runner
class ExperimentRunner(blueprints)[source]#

Bases: object

Description
Runs experiments as described by blueprints.
Parameters:

blueprints (Iterable[Blueprint]) – Blueprints describing experiments to run

register_evaluators(self, experiment) None[source]#
Description
Registers the experiment in all own evaluators.
Parameters:

experiment (Experiment) – Experiment to be registered

run_experiment(blueprint) None[source]#
Description
Runs the experiment based on blueprint on a single core.
Parameters:

blueprint (Blueprint) – The blueprint for the experiment

run_experiment_parallel(self, blueprint, nr_processes) None[source]#
Description
Runs a single experiment on multiple cores.
Parameters:
  • blueprint (Blueprint) – The blueprint for the experiment

  • nr_processes (int) – How many processes to run in parallel (default= 1/2 CPU count)

run_experiments(self, blueprints) None[source]#
Description
Runs multiple experiments on asingle core.
Parameters:

blueprints (Iterable[Blueprint]) – Blueprints for experiments (default= None)

run_experiments_parallel(self, blueprint, nr_processes) None[source]#
Description
Runs a multiple experiments. Runs either each experiment on multiple cores or one experiment per core.
Parameters:
  • blueprint (Iterbale[Blueprint]) – The blueprints for the experiments (default= None)

  • nr_processes (int) – How many processes to run in parallel (default= 1/2 CPU count)

  • parallel_sub_exp (bool) – Whether to run a single experiment in parallel (or multiple experiments in parallel) (default= True)

run_repetition(self, exp_config) None[source]#
Description
Runs the configured experiment once with the given exp_nr, ignoring the configured amount of repetitions.
Parameters:

exp_config (Tuple[Blueprint, int]) – A blueprint and a given experiment number