/*
* Encog(tm) Core v3.0 - Java Version
* http://www.heatonresearch.com/encog/
* http://code.google.com/p/encog-java/
* Copyright 2008-2011 Heaton Research, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
* For more information on Heaton Research copyrights, licenses
* and trademarks visit:
* http://www.heatonresearch.com/copyright
*/
package org.encog.mathutil.randomize;
import org.encog.EncogError;
import org.encog.ml.MLMethod;
import org.encog.neural.networks.BasicNetwork;
/**
* Implementation of <i>Nguyen-Widrow</i> weight initialization. This is the
* default weight initialization used by Encog, as it generally provides the
* most trainable neural network.
*
*
* @author St?phan Corriveau
*
*/
public class NguyenWidrowRandomizer extends RangeRandomizer implements
Randomizer {
private int inputCount;
private double beta;
/**
* Construct a Nguyen-Widrow randomizer.
*
* @param min
* The min of the range.
* @param max
* The max of the range.
*/
public NguyenWidrowRandomizer(final double min, final double max) {
super(min, max);
}
/**
* The <i>Nguyen-Widrow</i> initialization algorithm is the following :
* <br>
* 1. Initialize all weight of hidden layers with (ranged) random values<br>
* 2. For each hidden layer<br>
* 2.1 calculate beta value, 0.7 * Nth(#neurons of input layer) root of
* #neurons of current layer <br>
* 2.2 for each synapse<br>
* 2.1.1 for each weight <br>
* 2.1.2 Adjust weight by dividing by norm of weight for neuron and
* multiplying by beta value
* @param method The network to randomize.
*/
@Override
public final void randomize(final MLMethod method) {
if( !(method instanceof BasicNetwork) ) {
throw new EncogError("Ngyyen Widrow only works on BasicNetwork.");
}
BasicNetwork network = (BasicNetwork)method;
new RangeRandomizer(getMin(), getMax()).randomize(network);
int hiddenNeurons = 0;
for(int i=1;i<network.getLayerCount()-1;i++)
{
hiddenNeurons+=network.getLayerNeuronCount(i);
}
// can't really do much, use regular randomization
if (hiddenNeurons < 1) {
return;
}
this.inputCount = network.getInputCount();
this.beta = 0.7 * Math.pow(hiddenNeurons, 1.0 / network.getInputCount());
super.randomize(network);
}
/**
* Randomize one level of a neural network.
* @param network The network to randomize
* @param fromLayer The from level to randomize.
*/
public void randomize(final BasicNetwork network, int fromLayer)
{
int fromCount = network.getLayerTotalNeuronCount(fromLayer);
int toCount = network.getLayerNeuronCount(fromLayer+1);
for(int toNeuron = 0; toNeuron<toCount; toNeuron++)
{
double n = 0.0;
for (int fromNeuron = 0; fromNeuron < fromCount; fromNeuron++)
{
double w = network.getWeight(fromLayer, fromNeuron, toNeuron);
n += w*w;
}
n = Math.sqrt(n);
for (int fromNeuron = 0; fromNeuron < fromCount; fromNeuron++)
{
double w = network.getWeight(fromLayer, fromNeuron, toNeuron);
w = beta * w / n;
network.setWeight(fromLayer, fromNeuron, toNeuron, w);
}
}
}
}