18.2.7 Setting a Custom Loss Function and Batch Generator (for Anomaly Detection)

It is possible to select different loss functions for the supervised model by providing a LossFunction object, and different batch generators by providing a BatchGenerator object. This is useful for applications such as Anomaly Detection, which can be cast into the standard supervised framework but require different loss functions and batch generators.

SupervisedGraphWise model can use the DevNetLoss and the StratifiedOversamplingBatchGenerator. The DevNetLoss takes confidence margin and the value the anomaly takes in the target property as the two parameters.

The following example assumes that the convLayerConfig has already been defined:

• JShell
• Java
• Python

opg4j> import oracle.pgx.config.mllib.loss.LossFunctions
opg4j> import oracle.pgx.config.mllib.batchgenerator.BatchGenerators
opg4j> var predictionLayerConfig = analyst.graphWisePredictionLayerConfigBuilder().
         setHiddenDimension(32).
         setActivationFunction(ActivationFunction.LINEAR).
         build()
opg4j> var model = analyst.supervisedGraphWiseModelBuilder().
         setVertexInputPropertyNames("vertex_features").
         setEdgeInputPropertyNames("edge_features").
         setVertexTargetPropertyName("labels").
         setConvLayerConfigs(convLayerConfig).
         setPredictionLayerConfigs(predictionLayerConfig).
         setLossFunction(LossFunctions.devNetLoss(5.0, true)).
         setBatchGenerator(BatchGenerators.STRATIFIED_OVERSAMPLING).
         build()

import oracle.pgx.config.mllib.loss.LossFunctions;
import oracle.pgx.config.mllib.batchgenerator.BatchGenerators;

GraphWisePredictionLayerConfig predictionLayerConfig = analyst.graphWisePredictionLayerConfigBuilder()
    .setHiddenDimension(32)
    .setActivationFunction(ActivationFunction.LINEAR)
    .build();

SupervisedGraphWiseModel model = analyst.supervisedGraphWiseModelBuilder()
    .setVertexInputPropertyNames("vertex_features")
    .setEdgeInputPropertyNames("edge_features")
    .setVertexTargetPropertyName("labels")
    .setConvLayerConfigs(convLayerConfig)
    .setPredictionLayerConfigs(predictionLayerConfig)
    .setLossFunction(LossFunctions.devNetLoss(5.0, true))
    .setBatchGenerator(BatchGenerators.STRATIFIED_OVERSAMPLING)
    .build();

from pypgx.api.mllib import DevNetLoss

pred_layer_config = dict(hidden_dim=32,
                         activation_fn='LINEAR')

pred_layer = analyst.graphwise_pred_layer_config(**pred_layer_config)

params = dict(vertex_target_property_name="labels",
              conv_layer_config=[conv_layer],
              pred_layer_config=[pred_layer],
              vertex_input_property_names=["vertex_features"],
              edge_input_property_names=["edge_features"],
              loss_fn=DevNetLoss(5.0, True),
              batch_gen='Stratified_Oversampling',
              seed=17)

model = analyst.supervised_graphwise_builder(**params)