org.apache.spark.ml.ann

Interface LayerModel

All Superinterfaces:

java.io.Serializable

public interface LayerModel
extends scala.Serializable

Trait that holds Layer weights (or parameters). Implements functions needed for forward propagation, computing delta and gradient. Can return weights in Vector format.

Method Summary

Modifier and Type	Method and Description
void	computePrevDelta(breeze.linalg.DenseMatrix<Object> delta, breeze.linalg.DenseMatrix<Object> output, breeze.linalg.DenseMatrix<Object> prevDelta) Computes the delta for back propagation.
void	eval(breeze.linalg.DenseMatrix<Object> data, breeze.linalg.DenseMatrix<Object> output) Evaluates the data (process the data through the layer).
void	grad(breeze.linalg.DenseMatrix<Object> delta, breeze.linalg.DenseMatrix<Object> input, breeze.linalg.DenseVector<Object> cumGrad) Computes the gradient.
breeze.linalg.DenseVector<Object>	weights()

Method Detail

computePrevDelta

void computePrevDelta(breeze.linalg.DenseMatrix<Object> delta,
                      breeze.linalg.DenseMatrix<Object> output,
                      breeze.linalg.DenseMatrix<Object> prevDelta)

Computes the delta for back propagation. Delta is allocated based on the size provided by the LayerModel implementation and the stack (batch) size. Developer is responsible for checking the size of prevDelta when writing to it.

Parameters:

delta - delta of this layer

output - output of this layer

prevDelta - the previous delta (modified in place)

eval

void eval(breeze.linalg.DenseMatrix<Object> data,
          breeze.linalg.DenseMatrix<Object> output)

Evaluates the data (process the data through the layer). Output is allocated based on the size provided by the LayerModel implementation and the stack (batch) size. Developer is responsible for checking the size of output when writing to it.

Parameters:

data - data

output - output (modified in place)

grad

void grad(breeze.linalg.DenseMatrix<Object> delta,
          breeze.linalg.DenseMatrix<Object> input,
          breeze.linalg.DenseVector<Object> cumGrad)

Computes the gradient. cumGrad is a wrapper on the part of the weight vector. Size of cumGrad is based on weightSize provided by implementation of LayerModel.

Parameters:

delta - delta for this layer

input - input data

cumGrad - cumulative gradient (modified in place)

weights

breeze.linalg.DenseVector<Object> weights()