/*
* Encog(tm) Core v3.4 - Java Version
* http://www.heatonresearch.com/encog/
* https://github.com/encog/encog-java-core
* Copyright 2008-2016 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.neural.pnn;
import org.encog.ml.MLClassification;
import org.encog.ml.MLError;
import org.encog.ml.MLRegression;
import org.encog.ml.data.MLData;
import org.encog.ml.data.MLDataPair;
import org.encog.ml.data.MLDataSet;
import org.encog.ml.data.basic.BasicMLData;
import org.encog.ml.data.basic.BasicMLDataSet;
import org.encog.neural.NeuralNetworkError;
import org.encog.util.EngineArray;
import org.encog.util.simple.EncogUtility;
/**
* This class implements either a:
*
* Probabilistic Neural Network (PNN)
*
* General Regression Neural Network (GRNN)
*
* To use a PNN specify an output mode of classification, to make use of a GRNN
* specify either an output mode of regression or un-supervised autoassociation.
*
* The PNN/GRNN networks are potentially very useful. They share some
* similarities with RBF-neural networks and also the Support Vector Machine
* (SVM). These network types directly support the use of classification.
*
* The following book was very helpful in implementing PNN/GRNN's in Encog.
*
* Advanced Algorithms for Neural Networks: A C++ Sourcebook
*
* by Timothy Masters, PhD (<a href="http://www.timothymasters.info/"></a>) John Wiley and Sons
* Inc (Computers); April 3, 1995, ISBN: 0471105880
*/
public class BasicPNN extends AbstractPNN implements MLRegression, MLError, MLClassification {
/**
*
*/
private static final long serialVersionUID = -7990707837655024635L;
/**
* The sigma's specify the widths of each kernel used.
*/
private final double[] sigma;
/**
* The training samples that form the memory of this network.
*/
private BasicMLDataSet samples;
/**
* Used for classification, the number of cases in each class.
*/
private int[] countPer;
/**
* The prior probability weights.
*/
private double[] priors;
/**
* Construct a BasicPNN network.
*
* @param kernel
* The kernel to use.
* @param outmodel
* The output model for this network.
* @param inputCount
* The number of inputs in this network.
* @param outputCount
* The number of outputs in this network.
*/
public BasicPNN(final PNNKernelType kernel, final PNNOutputMode outmodel,
final int inputCount, final int outputCount) {
super(kernel, outmodel, inputCount, outputCount);
setSeparateClass(false);
this.sigma = new double[inputCount];
}
/**
* Compute the output from this network.
*
* @param input
* The input to the network.
* @return The output from the network.
*/
@Override
public MLData compute(final MLData input) {
final double[] out = new double[getOutputCount()];
double psum = 0.0;
int r = -1;
for (final MLDataPair pair : this.samples) {
r++;
if (r == getExclude()) {
continue;
}
double dist = 0.0;
for (int i = 0; i < getInputCount(); i++) {
double diff = input.getData(i) - pair.getInput().getData(i);
diff /= this.sigma[i];
dist += diff * diff;
}
if (getKernel() == PNNKernelType.Gaussian) {
dist = Math.exp(-dist);
} else if (getKernel() == PNNKernelType.Reciprocal) {
dist = 1.0 / (1.0 + dist);
}
if (dist < 1.e-40) {
dist = 1.e-40;
}
if (getOutputMode() == PNNOutputMode.Classification) {
final int pop = (int) pair.getIdeal().getData(0);
out[pop] += dist;
} else if (getOutputMode() == PNNOutputMode.Unsupervised) {
for (int i = 0; i < getInputCount(); i++) {
out[i] += dist * pair.getInput().getData(i);
}
psum += dist;
} else if (getOutputMode() == PNNOutputMode.Regression) {
for (int i = 0; i < getOutputCount(); i++) {
out[i] += dist * pair.getIdeal().getData(i);
}
psum += dist;
}
}
if (getOutputMode() == PNNOutputMode.Classification) {
psum = 0.0;
for (int i = 0; i < getOutputCount(); i++) {
if (this.priors[i] >= 0.0) {
out[i] *= this.priors[i] / this.countPer[i];
}
psum += out[i];
}
if (psum < 1.e-40) {
psum = 1.e-40;
}
for (int i = 0; i < getOutputCount(); i++) {
out[i] /= psum;
}
} else if (getOutputMode() == PNNOutputMode.Unsupervised) {
for (int i = 0; i < getInputCount(); i++) {
out[i] /= psum;
}
} else if (getOutputMode() == PNNOutputMode.Regression) {
for (int i = 0; i < getOutputCount(); i++) {
out[i] /= psum;
}
}
return new BasicMLData(out);
}
/**
* @return the countPer
*/
public int[] getCountPer() {
return this.countPer;
}
/**
* @return the priors
*/
public double[] getPriors() {
return this.priors;
}
/**
* @return the samples
*/
public BasicMLDataSet getSamples() {
return this.samples;
}
/**
* @return the sigma
*/
public double[] getSigma() {
return this.sigma;
}
/**
* @param samples
* the samples to set
*/
public void setSamples(final BasicMLDataSet samples) {
this.samples = samples;
// update counts per
if (getOutputMode() == PNNOutputMode.Classification) {
this.countPer = new int[getOutputCount()];
this.priors = new double[getOutputCount()];
for (final MLDataPair pair : samples) {
final int i = (int) pair.getIdeal().getData(0);
if (i >= this.countPer.length) {
throw new NeuralNetworkError(
"Training data contains more classes than neural network has output neurons to hold.");
}
this.countPer[i]++;
}
for (int i = 0; i < this.priors.length; i++) {
this.priors[i] = -1;
}
}
}
/**
* {@inheritDoc}
*/
@Override
public void updateProperties() {
// unneeded
}
/**
* {@inheritDoc}
*/
@Override
public double calculateError(MLDataSet data) {
if (getOutputMode() == PNNOutputMode.Classification) {
return EncogUtility.calculateClassificationError(this, data);
} else {
return EncogUtility.calculateRegressionError(this, data);
}
}
/**
* {@inheritDoc}
*/
@Override
public int classify(MLData input) {
MLData output = compute(input);
return EngineArray.maxIndex(output.getData());
}
}