org.apache.spark.ml.ann

Interface TopologyModel

All Superinterfaces:

java.io.Serializable

public interface TopologyModel
extends scala.Serializable

Trait for ANN topology model

Method Summary

Modifier and Type	Method and Description
double	computeGradient(breeze.linalg.DenseMatrix<Object> data, breeze.linalg.DenseMatrix<Object> target, Vector cumGradient, int blockSize) Computes gradient for the network
breeze.linalg.DenseMatrix<Object>[]	forward(breeze.linalg.DenseMatrix<Object> data, boolean includeLastLayer) Forward propagation
LayerModel[]	layerModels() Array of layer models
Layer[]	layers() Array of layers
Vector	predict(Vector features) Prediction of the model.
Vector	predictRaw(Vector features) Raw prediction of the model.
Vector	raw2ProbabilityInPlace(Vector rawPrediction) Probability of the model.
Vector	weights()

Method Detail

computeGradient

double computeGradient(breeze.linalg.DenseMatrix<Object> data,
                       breeze.linalg.DenseMatrix<Object> target,
                       Vector cumGradient,
                       int blockSize)

Computes gradient for the network

Parameters:

data - input data

target - target output

cumGradient - cumulative gradient

blockSize - block size

Returns:

error

forward

breeze.linalg.DenseMatrix<Object>[] forward(breeze.linalg.DenseMatrix<Object> data,
                                            boolean includeLastLayer)

Forward propagation

Parameters:

data - input data

includeLastLayer - Include the last layer in the output. In MultilayerPerceptronClassifier, the last layer is always softmax; the last layer of outputs is needed for class predictions, but not for rawPrediction.

Returns:

array of outputs for each of the layers

layerModels

LayerModel[] layerModels()

Array of layer models

Returns:

(undocumented)

layers

Layer[] layers()

Array of layers

Returns:

(undocumented)

predict

Vector predict(Vector features)

Prediction of the model. See ProbabilisticClassificationModel

Parameters:

features - input features

Returns:

prediction

predictRaw

Vector predictRaw(Vector features)

Raw prediction of the model. See ProbabilisticClassificationModel

Parameters:

features - input features

Returns:

raw prediction

Note: This interface is only used for classification Model.

raw2ProbabilityInPlace

Vector raw2ProbabilityInPlace(Vector rawPrediction)

Probability of the model. See ProbabilisticClassificationModel

Parameters:

rawPrediction - raw prediction vector

Returns:

probability

Note: This interface is only used for classification Model.

weights

Vector weights()