oracle.pgx.api.mllib

Class SupervisedEdgeWiseModelBuilder

java.lang.Object

oracle.pgx.api.mllib.WiseModelBuilder<Config,Self>

oracle.pgx.api.mllib.EdgeWiseModelBuilder<SupervisedEdgeWiseModel,SupervisedEdgeWiseModelConfig,SupervisedEdgeWiseModelBuilder>

oracle.pgx.api.mllib.SupervisedEdgeWiseModelBuilder

public class SupervisedEdgeWiseModelBuilder
extends EdgeWiseModelBuilder<SupervisedEdgeWiseModel,SupervisedEdgeWiseModelConfig,SupervisedEdgeWiseModelBuilder>

Builder for SupervisedEdgeWiseModel The builder can be used to set the configuration of the model and create the model object.

Since:

23.1

Constructor Summary

Constructors

Constructor and Description
SupervisedEdgeWiseModelBuilder(PgxSession session, oracle.pgx.api.internal.Core core, java.util.function.Supplier<java.lang.String> keystorePathSupplier, java.util.function.Supplier<char[]> keystorePasswordSupplier, java.util.function.BiFunction<PgxSession,oracle.pgx.api.internal.Graph,PgxGraph> graphConstructor)

Method Summary

Modifier and Type	Method and Description
SupervisedEdgeWiseModel	build() Builds the SupervisedEdgeWise model with the specified parameters.
SupervisedEdgeWiseModelBuilder	setBatchGenerator(BatchGenerator batchGenerator) Sets the batch generator.
SupervisedEdgeWiseModelBuilder	setClassWeights(java.util.Map<?,java.lang.Float> classWeights) Set the class weights to be used in the loss function.
SupervisedEdgeWiseModelBuilder	setEdgeCombinationMethod(EdgeCombinationMethod edgeCombinationMethod) Sets the edge embedding method for the algorithm.
SupervisedEdgeWiseModelBuilder	setEdgeTargetPropertyName(java.lang.String propertyName) Set the target (labels) for the algorithm in the form of a property name of the graph.
SupervisedEdgeWiseModelBuilder	setLossFunction(LossFunction lossFunction) Sets the loss function for the algorithm.
SupervisedEdgeWiseModelBuilder	setNormalize(boolean normalize) Whether to l2 normalize the output of the convolutional layers.
SupervisedEdgeWiseModelBuilder	setPredictionLayerConfigs(GraphWisePredictionLayerConfig... layerConfigs) Set the prediction layer configurations (See GraphWisePredictionLayerConfig.

Methods inherited from class oracle.pgx.api.mllib.EdgeWiseModelBuilder

setTargetEdgeLabels, setTargetEdgeLabels

Methods inherited from class oracle.pgx.api.mllib.WiseModelBuilder

setBatchSize, setConvLayerConfigs, setEdgeInputPropertyNames, setEdgeInputPropertyNames, setEmbeddingDim, setLearningRate, setNumEpochs, setSeed, setShuffle, setStandardize, setVertexInputPropertyNames, setVertexInputPropertyNames, setWeightDecay

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

SupervisedEdgeWiseModelBuilder

public SupervisedEdgeWiseModelBuilder(PgxSession session,
                                      oracle.pgx.api.internal.Core core,
                                      java.util.function.Supplier<java.lang.String> keystorePathSupplier,
                                      java.util.function.Supplier<char[]> keystorePasswordSupplier,
                                      java.util.function.BiFunction<PgxSession,oracle.pgx.api.internal.Graph,PgxGraph> graphConstructor)

Method Detail

build

public SupervisedEdgeWiseModel build()
                              throws java.lang.InterruptedException,
                                     java.util.concurrent.ExecutionException

Builds the SupervisedEdgeWise model with the specified parameters.

Specified by:

build in class EdgeWiseModelBuilder<SupervisedEdgeWiseModel,SupervisedEdgeWiseModelConfig,SupervisedEdgeWiseModelBuilder>

Returns:

SupervisedEdgeWise model

Throws:

java.lang.InterruptedException

java.util.concurrent.ExecutionException

Since:

23.1

setBatchGenerator

public SupervisedEdgeWiseModelBuilder setBatchGenerator(BatchGenerator batchGenerator)

Sets the batch generator. See BatchGenerator

default: StandardBatchGenerator

Parameters:

batchGenerator - batch generator

Returns:

this

Since:

23.1

setClassWeights

public SupervisedEdgeWiseModelBuilder setClassWeights(java.util.Map<?,java.lang.Float> classWeights)

Set the class weights to be used in the loss function. The loss for the corresponding class will be multiplied by the factor given in this map. If null, uniform class weights will be used.

default: SupervisedEdgeWiseModelConfig.DEFAULT_CLASS_MAP

Parameters:

classWeights - map from classes to weights

Returns:

this

Since:

23.1

setEdgeCombinationMethod

public SupervisedEdgeWiseModelBuilder setEdgeCombinationMethod(EdgeCombinationMethod edgeCombinationMethod)

Sets the edge embedding method for the algorithm. See EdgeCombinationMethod

default: EdgeCombinationMethods.DEFAULT_CONCAT_METHOD

Parameters:

edgeCombinationMethod - edge embedding method

Returns:

this

Since:

23.1

setEdgeTargetPropertyName

public SupervisedEdgeWiseModelBuilder setEdgeTargetPropertyName(java.lang.String propertyName)

Set the target (labels) for the algorithm in the form of a property name of the graph.

Supported property types are given by SupervisedEdgeWiseModelConfig.SUPPORTED_LABEL_TYPES.

Parameters:

propertyName - property name

Returns:

this

Since:

23.1

setLossFunction

public SupervisedEdgeWiseModelBuilder setLossFunction(LossFunction lossFunction)

Sets the loss function for the algorithm. See LossFunction

default: SoftmaxCrossEntropyLoss

Parameters:

lossFunction - loss function

Returns:

this

Since:

23.1

setNormalize

public SupervisedEdgeWiseModelBuilder setNormalize(boolean normalize)

Whether to l2 normalize the output of the convolutional layers.

default: SupervisedEdgeWiseModelConfig.DEFAULT_NORMALIZE

Parameters:

normalize - whether to normalize

Returns:

this

Since:

23.1

setPredictionLayerConfigs

public SupervisedEdgeWiseModelBuilder setPredictionLayerConfigs(GraphWisePredictionLayerConfig... layerConfigs)

Set the prediction layer configurations (See GraphWisePredictionLayerConfig. You must pass at least one prediction layer config (or leave it as default).

Note that for the last layer:

• the activation function will be ignored (and replaced with the activation function of the loss function, e.g. softmax or sigmoid).
• the hidden dimension will be ignored (and replaced with the number of classes)

default: SupervisedEdgeWiseModelConfig.DEFAULT_PREDICTION_LAYER_CONFIGS

Parameters:

layerConfigs - layer configs

Returns:

this

Since:

23.1