Module datarush.library
Package com.pervasive.datarush.ports.model
Provides implementations of port objects dealing with the flow of single objects
between operators. Typically, these objects represent some abstract model
(such as a prediction model, for example PMML) instead of a set of records.
Thus ports which transfer single objects are referred to as model ports
within DataRush.
Key points to be aware of when using model ports:
- Models are assumed to be small; that is, small enough to fit into memory.
- Partial models may be constructed in parallel and then reduced to a single model. When this is the case, a special merge handler must be provided to combine the models.
- Consumers of models may share the same model object. Therefore care should be taken that models not be mutated by reader.
-
Interface Summary Interface Description ModelMergeHandler<T> A model merge handle is responsible for merging partial models into a final model.ModelStorageHandler<T> A model storage handler is responsible for handling the persistence of a model object. -
Class Summary Class Description AbstractModelPort<T> Common base class for all types of model ports.AbstractModelPortFactory<T,P extends AbstractModelPort<T>> Base class for model port factories.AbstractModelPortMetadata Base class for model metadata.DoneSignalPort Used to coordinate processing between operators that have a processing dependency but no explicit data dependency.SerializableModelStorageHandler<T extends Serializable> ModelStorageHandler that supports persistence of any Java-serializable model object.SimpleModelPort<T extends Serializable> A simple implementation for model ports.SimpleModelPort.Factory<T extends Serializable> Factory for creating SimpleModelPorts