/*
* Encog(tm) Core v3.3 - Java Version
* http://www.heatonresearch.com/encog/
* https://github.com/encog/encog-java-core
* Copyright 2008-2014 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.networks.training.lma;
import org.encog.mathutil.error.ErrorCalculation;
import org.encog.mathutil.matrices.decomposition.LUDecomposition;
import org.encog.mathutil.matrices.hessian.ComputeHessian;
import org.encog.mathutil.matrices.hessian.HessianCR;
import org.encog.ml.MLMethod;
import org.encog.ml.TrainingImplementationType;
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.BasicMLDataPair;
import org.encog.ml.train.BasicTraining;
import org.encog.neural.networks.BasicNetwork;
import org.encog.neural.networks.structure.NetworkCODEC;
import org.encog.neural.networks.training.TrainingError;
import org.encog.neural.networks.training.propagation.TrainingContinuation;
import org.encog.util.concurrency.MultiThreadable;
import org.encog.util.validate.ValidateNetwork;
/**
* Trains a neural network using a Levenberg Marquardt algorithm (LMA). This
* training technique is based on the mathematical technique of the same name.
*
* The LMA interpolates between the Gauss-Newton algorithm (GNA) and the
* method of gradient descent (similar to what is used by backpropagation.
* The lambda parameter determines the degree to which GNA and Gradient
* Descent are used. A lower lambda results in heavier use of GNA,
* whereas a higher lambda results in a heavier use of gradient descent.
*
* Each iteration starts with a low lambda that builds if the improvement
* to the neural network is not desirable. At some point the lambda is
* high enough that the training method reverts totally to gradient descent.
*
* This allows the neural network to be trained effectively in cases where GNA
* provides the optimal training time, but has the ability to fall back to the
* more primitive gradient descent method
*
* LMA finds only a local minimum, not a global minimum.
*
* References:
* http://www.heatonresearch.com/wiki/LMA
* http://en.wikipedia.org/wiki/Levenberg%E2%80%93Marquardt_algorithm
* http://en.wikipedia.org/wiki/Finite_difference_method
* http://crsouza.blogspot.com/2009/11/neural-network-learning-by-levenberg_18.html
* http://mathworld.wolfram.com/FiniteDifference.html
* http://www-alg.ist.hokudai.ac.jp/~jan/alpha.pdf -
* http://www.inference.phy.cam.ac.uk/mackay/Bayes_FAQ.html
*
*/
public class LevenbergMarquardtTraining extends BasicTraining implements MultiThreadable {
/**
* The amount to scale the lambda by.
*/
public static final double SCALE_LAMBDA = 10.0;
/**
* The max amount for the LAMBDA.
*/
public static final double LAMBDA_MAX = 1e25;
/**
* Utility class to compute the Hessian.
*/
private ComputeHessian hessian;
/**
* The network that is to be trained.
*/
private final BasicNetwork network;
/**
* The training set that we are using to train.
*/
private final MLDataSet indexableTraining;
/**
* The training set length.
*/
private final int trainingLength;
/**
* How many weights are we dealing with?
*/
private final int weightCount;
/**
* The neural network weights and bias values.
*/
private double[] weights;
/**
* The lambda, or damping factor. This is increased until a desirable
* adjustment is found.
*/
private double lambda;
/**
* The diagonal of the hessian.
*/
private final double[] diagonal;
/**
* The amount to change the weights by.
*/
private double[] deltas;
/**
* The training elements.
*/
private final MLDataPair pair;
/**
* Is the init complete?
*/
private boolean initComplete;
/**
* Construct the LMA object.
*
* @param network
* The network to train. Must have a single output neuron.
* @param training
* The training data to use. Must be indexable.
*/
public LevenbergMarquardtTraining(final BasicNetwork network,
final MLDataSet training) {
this(network,training,new HessianCR());
}
/**
* Construct the LMA object.
*
* @param network
* The network to train. Must have a single output neuron.
* @param training
* The training data to use. Must be indexable.
*/
public LevenbergMarquardtTraining(final BasicNetwork network,
final MLDataSet training, final ComputeHessian h) {
super(TrainingImplementationType.Iterative);
ValidateNetwork.validateMethodToData(network, training);
setTraining(training);
this.indexableTraining = getTraining();
this.network = network;
this.trainingLength = (int) this.indexableTraining.getRecordCount();
this.weightCount = this.network.getStructure().calculateSize();
this.lambda = 0.1;
this.deltas = new double[this.weightCount];
this.diagonal = new double[this.weightCount];
final BasicMLData input = new BasicMLData(
this.indexableTraining.getInputSize());
final BasicMLData ideal = new BasicMLData(
this.indexableTraining.getIdealSize());
this.pair = new BasicMLDataPair(input, ideal);
this.hessian = h;
}
private void saveDiagonal() {
double[][] h = this.hessian.getHessian();
for (int i = 0; i < this.weightCount; i++) {
this.diagonal[i] = h[i][i];
}
}
@Override
public boolean canContinue() {
return false;
}
/**
* @return The trained network.
*/
@Override
public MLMethod getMethod() {
return this.network;
}
/**
* @return The SSE error with the current weights.
*/
private double calculateError() {
ErrorCalculation result = new ErrorCalculation();
for (int i = 0; i < this.trainingLength; i++) {
this.indexableTraining.getRecord(i, this.pair);
final MLData actual = this.network.compute(this.pair.getInput());
result.updateError(actual.getData(), this.pair.getIdeal().getData(),pair.getSignificance());
}
return result.calculateESS();
}
private void applyLambda() {
double[][] h = this.hessian.getHessian();
for (int i = 0; i < this.weightCount; i++) {
h[i][i] = this.diagonal[i] + this.lambda;
}
}
/**
* Perform one iteration.
*/
@Override
public void iteration() {
if( !initComplete ) {
this.hessian.init(network, getTraining());
this.initComplete = true;
}
LUDecomposition decomposition = null;
preIteration();
this.hessian.clear();
this.weights = NetworkCODEC.networkToArray(this.network);
this.hessian.compute();
double currentError = this.hessian.getSSE();
saveDiagonal();
final double startingError = currentError;
boolean done = false;
boolean singular;
while (!done) {
applyLambda();
decomposition = new LUDecomposition(this.hessian.getHessianMatrix());
singular = decomposition.isNonsingular();
if (singular) {
this.deltas = decomposition.Solve(this.hessian.getGradients());
updateWeights();
currentError = calculateError();
}
if ( !singular || currentError >= startingError) {
this.lambda *= LevenbergMarquardtTraining.SCALE_LAMBDA;
if( this.lambda> LevenbergMarquardtTraining.LAMBDA_MAX ) {
this.lambda = LevenbergMarquardtTraining.LAMBDA_MAX;
done = true;
}
} else {
this.lambda /= LevenbergMarquardtTraining.SCALE_LAMBDA;
done = true;
}
}
setError(currentError);
postIteration();
}
/**
* {@inheritDoc}
*/
@Override
public TrainingContinuation pause() {
return null;
}
/**
* {@inheritDoc}
*/
@Override
public void resume(final TrainingContinuation state) {
}
/**
* Update the weights in the neural network.
*/
public void updateWeights() {
final double[] w = this.weights.clone();
for (int i = 0; i < w.length; i++) {
w[i] += this.deltas[i];
}
NetworkCODEC.arrayToNetwork(w, this.network);
}
/**
* @return The Hessian calculation method used.
*/
public ComputeHessian getHessian() {
return hessian;
}
@Override
public int getThreadCount() {
if( this.hessian instanceof MultiThreadable ) {
return ((MultiThreadable)this.hessian).getThreadCount();
} else {
return 1;
}
}
@Override
public void setThreadCount(int numThreads) {
if( this.hessian instanceof MultiThreadable ) {
((MultiThreadable)this.hessian).setThreadCount(numThreads);
} else if(numThreads!=1 && numThreads!=0) {
throw new TrainingError("The Hessian object in use("+this.hessian.getClass().toString()+") does not support multi-threaded mode.");
}
}
}