Building an Unsupervised GraphWise Model Using Partitioned Graphs

17.4.5 Building an Unsupervised GraphWise Model Using Partitioned Graphs

You can build an Unsupervised GraphWise model using partitioned graphs which have different providers and features.

opg4j> analyst.unsupervisedGraphWiseModelBuilder().
        setVertexInputPropertyNames("vertex_provider1_features", "vertex_provider2_features").
        setEdgeInputPropertyNames("edge_provider_features").
        setVertexTargetPropertyName("target_property").
        build()

UnsupervisedGraphWiseModel model = analyst.unsupervisedGraphWiseModelBuilder()
    .setVertexInputPropertyNames("vertex_provider1_features", "vertex_provider2_features")
    .setEdgeInputPropertyNames("edge_provider_features")
    .setVertexTargetPropertyName("target_property")
    .build();

params = dict(vertex_input_property_names=["vertex_provider1_features", "vertex_provider2_features"],
              edge_input_property_names=["edge_provider_features"])
model = analyst.unsupervised_graphwise_builder(**params)

Also, you can select specific providers as shown:

opg4j> var model = analyst.unsupervisedGraphWiseModelBuilder().
         setVertexInputPropertyNames("vertex_provider1_features", "vertex_provider2_features").
         setEdgeInputPropertyNames("edge_provider_features").
         setTargetVertexLabels("provider1").
         build()

UnsupervisedGraphWiseModel model = analyst.unsupervisedGraphWiseModelBuilder()
    .setVertexInputPropertyNames("vertex_provider1_features", "vertex_provider2_features")
    .setEdgeInputPropertyNames("edge_provider_features")
    .setTargetVertexLabels("provider1")
    .build();

params = dict(vertex_input_property_names=["vertex_provider1_features", "vertex_provider2_features"],
              edge_input_property_names=["edge_provider_features"],
              target_vertex_labels=["provider1"])
model = analyst.unsupervised_graphwise_builder(**params)

If you wish to control the flow of the embeddings at each layer, you can enable or disable the required connections. By default, all the connections are enabled.

opg4j> var convLayerConfig = analyst.graphWiseConvLayerConfigBuilder().
         setNumSampledNeighbors(25).
         useVertexToVertexConnection(true).
         useEdgeToVertexConnection(true).
         useEdgeToEdgeConnection(false).
         useVertexToEdgeConnection(false).
         build()
opg4j> var model = analyst.unsupervisedGraphWiseModelBuilder().
         setVertexInputPropertyNames("vertex_provider1_features", "vertex_provider2_features").
         setEdgeInputPropertyNames("edge_provider_features").
         setTargetVertexLabels("provider1").
         build()

GraphWiseConvLayerConfig convLayerConfig = analyst.graphWiseConvLayerConfigBuilder()
    .setNumSampledNeighbors(10)
    .useVertexToVertexConnection(true)
    .useEdgeToVertexConnection(true)
    .useEdgeToEdgeConnection(false)
    .useVertexToEdgeConnection(false)
    .build();

UnsupervisedGraphWiseModel model = analyst.unsupervisedGraphWiseModelBuilder()
    .setVertexInputPropertyNames("vertex_provider1_features", "vertex_provider2_features")
    .setEdgeInputPropertyNames("edge_provider_features")
    .setTargetVertexLabels("provider1")
    .setConvLayerConfigs(convLayerConfig)
    .build();

conv_layer_config = dict(num_sampled_neighbors=25,
                         activation_fn='tanh',
                         weight_init_scheme='xavier',
                         neighbor_weight_property_name=weightProperty,
                         vertex_to_vertex_connection=True,
                         edge_to_vertex_connection=True,
                         vertex_to_edge_connection=False,
                         edge_to_edge_connection=False)

conv_layer = analyst.graphwise_conv_layer_config(**conv_layer_config)

params = dict(vertex_input_property_names=["vertex_provider1_features", "vertex_provider2_features"],
              edge_input_property_names=["edge_provider_features"],
              target_vertex_labels=["provider1"],
              conv_layer_config=[conv_layer])

model = analyst.unsupervised_graphwise_builder(**params)

Parent topic: Using the Unsupervised GraphWise Algorithm (Vertex Embeddings)