reagent.publishers package

Submodules

reagent.publishers.file_system_publisher module

class reagent.publishers.file_system_publisher.FileSystemPublisher

Bases: object

reagent.publishers.model_publisher module

class reagent.publishers.model_publisher.ModelPublisher

Bases: object

Base class for model publisher. All publishers should subclass from this so that they can be registered in the workflows.

REGISTRY = {}

REGISTRY_FROZEN = False

REGISTRY_NAME = 'ModelPublisher'

abstract do_publish(model_manager: reagent.model_managers.model_manager.ModelManager, training_output: reagent.workflow.types.RLTrainingOutput, setup_data: Optional[Dict[str, bytes]], recurring_workflow_ids: Dict[str, int], child_workflow_id: int, recurring_period: Optional[datetime.datetime]) → reagent.core.result_registries.PublishingResult

This method takes RLTrainingOutput so that it can extract anything it might need from it.

ModelManager is given here so that config can be shared

publish(model_manager: reagent.model_managers.model_manager.ModelManager, training_output: reagent.workflow.types.RLTrainingOutput, setup_data: Optional[Dict[str, bytes]], recurring_workflow_ids: Dict[str, int], child_workflow_id: int, recurring_period: Optional[datetime.datetime])

This method takes RLTrainingOutput so that it can extract anything it might need from it.

ModelManager is given here so that config can be shared

reagent.publishers.no_publishing module

class reagent.publishers.no_publishing.NoPublishing

Bases: reagent.publishers.model_publisher.ModelPublisher

This is an example of how to create a publisher. This publisher performs no publishing. In your own publisher, you would want to have validate() performs some publishing.

do_publish(model_manager: reagent.model_managers.model_manager.ModelManager, training_output: reagent.workflow.types.RLTrainingOutput, setup_data: Optional[Dict[str, bytes]], recurring_workflow_ids: Dict[str, int], child_workflow_id: int, recurring_period: Optional[datetime.datetime]) → reagent.core.result_types.NoPublishingResults

This method takes RLTrainingOutput so that it can extract anything it might need from it.

ModelManager is given here so that config can be shared

reagent.publishers.union module

class reagent.publishers.union.ModelPublisher__Union(NoPublishing: Optional[reagent.publishers.no_publishing.NoPublishing] = None)

Bases: reagent.core.tagged_union.TaggedUnion

NoPublishing: Optional[reagent.publishers.no_publishing.NoPublishing] = None

make_union_instance(instance_class=None)

Module contents