hex.grid

Interface HyperSpaceWalker.HyperSpaceIterator<MP extends Model.Parameters>

Enclosing interface:

HyperSpaceWalker<MP extends Model.Parameters,C extends HyperSpaceSearchCriteria>

public static interface HyperSpaceWalker.HyperSpaceIterator<MP extends Model.Parameters>

Method Summary

Modifier and Type	Method and Description
boolean	hasNext(Model previousModel) Returns true if the iterator can continue.
MP	nextModelParameters(Model previousModel) Get next model parameters.
void	onModelFailure(Model failedModel, java.util.function.Consumer<java.lang.Object[]> withFailedModelHyperParams) Inform the Iterator that a model build failed in case it needs to adjust its internal state.

Method Detail

nextModelParameters

MP nextModelParameters(Model previousModel)

Get next model parameters.

It should return model parameters for next point in hyper space. Throws NoSuchElementException if there is no remaining point in space to explore.

The method can optimize based on previousModel, but should be able to handle null-value.

Parameters:

previousModel - model generated for the previous point in hyper space, can be null.

Returns:

model parameters for next point in hyper space or null if there is no such point.

Throws:

java.lang.IllegalArgumentException - when model parameters cannot be constructed

java.util.NoSuchElementException - if the iteration has no more elements

hasNext

boolean hasNext(Model previousModel)

Returns true if the iterator can continue. Takes into account strategy-specific stopping criteria, if any.

Parameters:

previousModel - optional parameter which helps to determine next step, can be null

Returns:

true if the iterator can produce one more model parameters configuration.

onModelFailure

void onModelFailure(Model failedModel,
                    java.util.function.Consumer<java.lang.Object[]> withFailedModelHyperParams)

Inform the Iterator that a model build failed in case it needs to adjust its internal state. Implementations are expected to consume the withFailedModelHyperParams callback with the hyperParams used to create the failed model.

Parameters:

failedModel: - the model whose training failed.

withFailedModelHyperParams: - consumes the "raw" hyperparameters values used for the failed model.