Package ml.shifu.guagua.example.nn

Class Gradient

java.lang.Object

ml.shifu.guagua.example.nn.Gradient

public class Gradient
extends Object

Gradient is copied from Encog framework. The reason is that we original Gradient don't pop up gradients outside. While we need gradients accumulated into NNMaster to update NN weights.

Constructor Summary

Constructors

Constructor	Description
Gradient(org.encog.neural.flat.FlatNetwork theNetwork, org.encog.ml.data.MLDataSet theTraining, double[] flatSpot, org.encog.neural.error.ErrorFunction ef)	Construct a gradient worker.

Method Summary

Modifier and Type	Method	Description
double	getError()
org.encog.mathutil.error.ErrorCalculation	getErrorCalculation()
double[]	getGradients()
double[]	getLayerDelta()
org.encog.neural.flat.FlatNetwork	getNetwork()
double[]	getWeights()
void	run()	Perform the gradient calculation
void	setParams(org.encog.neural.networks.BasicNetwork network)
void	setWeights(double[] weights)

Methods inherited from class java.lang.Object

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

Constructor Detail

Gradient

public Gradient(org.encog.neural.flat.FlatNetwork theNetwork,
                org.encog.ml.data.MLDataSet theTraining,
                double[] flatSpot,
                org.encog.neural.error.ErrorFunction ef)

Construct a gradient worker.

Method Detail

run

public final void run()

Perform the gradient calculation

getErrorCalculation

public org.encog.mathutil.error.ErrorCalculation getErrorCalculation()

getGradients

public double[] getGradients()

Returns:

the gradients

getError

public double getError()

Returns:

the error

getWeights

public double[] getWeights()

Returns:

the weights

setWeights

public void setWeights(double[] weights)

Parameters:

weights - the weights to set

setParams

public void setParams(org.encog.neural.networks.BasicNetwork network)

getNetwork

public org.encog.neural.flat.FlatNetwork getNetwork()

getLayerDelta

public double[] getLayerDelta()