Execution

module crystallize.plugins.execution

Execution is delegated out of Experiment into two collaborators:

• An execution plugin (e.g., SerialExecution, ParallelExecution, or AsyncExecution) that drives the replicate loop via run_experiment_loop.
• ResultAggregator (from crystallize.experiments.aggregation) which collates replicate outputs, seeds, provenance, and metrics into a Result.

Experiment.run/arun selects the strategy plugin, passes in the replicate callable, and then hands the collected replicate results to ResultAggregator. Override or swap these collaborators instead of subclassing Experiment to change aggregation behavior; the legacy _aggregate_results hook is no longer part of the execution path.

Global Variables

• TYPE_CHECKING
• VALID_EXECUTOR_TYPES

---

class SerialExecution

Execute replicates one after another within the main process.

method SerialExecution.__init__

__init__(progress: 'bool' = False) → None

---

method SerialExecution.run_experiment_loop

run_experiment_loop(
    experiment: "'Experiment'",
    replicate_fn: 'Callable[[int], Any]'
) → List[Any]

---

class ParallelExecution

Run SYNC replicates concurrently using ThreadPoolExecutor or ProcessPoolExecutor.

method ParallelExecution.__init__

__init__(
    max_workers: 'Optional[int]' = None,
    executor_type: 'str' = 'thread',
    progress: 'bool' = False
) → None

---

method ParallelExecution.run_experiment_loop

run_experiment_loop(
    experiment: "'Experiment'",
    replicate_fn: 'Callable[[int], Any]'
) → List[Any]

---

class AsyncExecution

Run async replicates concurrently using asyncio.gather.

method AsyncExecution.__init__

__init__(progress: 'bool' = False) → None

---

method AsyncExecution.run_experiment_loop

run_experiment_loop(
    experiment: "'Experiment'",
    replicate_fn: 'Callable[[int], Any]'
) → List[Any]