/**
* Convert to an array. This is used with some training algorithms that require that the "memory"
* of the neuron(the weight and bias values) be expressed as a linear array.
*
* @param network The network to encode.
* @return The memory of the neuron.
*/
public static double[] networkToArray(final BasicNetwork network) {
final int size = network.getStructure().calculateSize();
// allocate an array to hold
final double[] result = new double[size];
int index = 0;
for (final Layer layer : network.getStructure().getLayers()) {
// process layer bias
if (layer.hasBias()) {
for (int i = 0; i < layer.getNeuronCount(); i++) {
result[index++] = layer.getBiasWeight(i);
}
}
// process synapses
for (final Synapse synapse : network.getStructure().getPreviousSynapses(layer)) {
if (synapse.getMatrix() != null) {
// process each weight matrix
for (int x = 0; x < synapse.getToNeuronCount(); x++) {
for (int y = 0; y < synapse.getFromNeuronCount(); y++) {
result[index++] = synapse.getMatrix().get(y, x);
}
}
}
}
}
return result;
}
public void trainAndSave() {
System.out.println("Training XOR network to under 1% error rate.");
BasicNetwork network = new BasicNetwork();
network.addLayer(new BasicLayer(2));
network.addLayer(new BasicLayer(2));
network.addLayer(new BasicLayer(1));
network.getStructure().finalizeStructure();
network.reset();
NeuralDataSet trainingSet = new BasicNeuralDataSet(XOR_INPUT, XOR_IDEAL);
// train the neural network
final Train train = new ResilientPropagation(network, trainingSet);
do {
train.iteration();
} while (train.getError() > 0.009);
double e = network.calculateError(trainingSet);
System.out.println("Network traiined to error: " + e);
System.out.println("Saving network");
final EncogPersistedCollection encog = new EncogPersistedCollection(FILENAME);
encog.create();
encog.add("network", network);
}
@Test
public void testDualOutput() {
BasicNetwork network = new BasicNetwork();
network.addLayer(new BasicLayer(null, true, 2));
network.addLayer(new BasicLayer(new ActivationSigmoid(), true, 2));
network.addLayer(new BasicLayer(new ActivationSigmoid(), false, 2));
network.getStructure().finalizeStructure();
(new ConsistentRandomizer(-1, 1)).randomize(network);
MLDataSet trainingData = new BasicMLDataSet(XOR.XOR_INPUT, XOR.XOR_IDEAL2);
HessianFD testFD = new HessianFD();
testFD.init(network, trainingData);
testFD.compute();
// dump(testFD, "FD");
HessianCR testCR = new HessianCR();
testCR.init(network, trainingData);
testCR.compute();
// dump(testCR, "CR");
Assert.assertTrue(testCR.getHessianMatrix().equals(testFD.getHessianMatrix(), 4));
}
/**
* Generate the network.
*
* @return The generated network.
*/
public BasicNetwork generate() {
Layer input, instar, outstar;
int y = PatternConst.START_Y;
final BasicNetwork network = new BasicNetwork();
network.addLayer(input = new BasicLayer(new ActivationLinear(), false, this.inputCount));
network.addLayer(instar = new BasicLayer(new ActivationCompetitive(), false, this.instarCount));
network.addLayer(outstar = new BasicLayer(new ActivationLinear(), false, this.outstarCount));
network.getStructure().finalizeStructure();
network.reset();
input.setX(PatternConst.START_X);
input.setY(y);
y += PatternConst.INC_Y;
instar.setX(PatternConst.START_X);
instar.setY(y);
y += PatternConst.INC_Y;
outstar.setX(PatternConst.START_X);
outstar.setY(y);
// tag as needed
network.tagLayer(BasicNetwork.TAG_INPUT, input);
network.tagLayer(BasicNetwork.TAG_OUTPUT, outstar);
network.tagLayer(CPNPattern.TAG_INSTAR, instar);
network.tagLayer(CPNPattern.TAG_OUTSTAR, outstar);
return network;
}
/**
* Create a feed forward network.
*
* @param architecture The architecture string to use.
* @param input The input count.
* @param output The output count.
* @return The feedforward network.
*/
public final MLMethod create(final String architecture, final int input, final int output) {
if (input <= 0) {
throw new EncogError("Must have at least one input for feedforward.");
}
if (output <= 0) {
throw new EncogError("Must have at least one output for feedforward.");
}
final BasicNetwork result = new BasicNetwork();
final List<String> layers = ArchitectureParse.parseLayers(architecture);
ActivationFunction af = new ActivationLinear();
int questionPhase = 0;
for (final String layerStr : layers) {
int defaultCount;
// determine default
if (questionPhase == 0) {
defaultCount = input;
} else {
defaultCount = output;
}
final ArchitectureLayer layer = ArchitectureParse.parseLayer(layerStr, defaultCount);
final boolean bias = layer.isBias();
String part = layer.getName();
if (part != null) {
part = part.trim();
} else {
part = "";
}
ActivationFunction lookup = this.factory.create(part);
if (lookup != null) {
af = lookup;
} else {
if (layer.isUsedDefault()) {
questionPhase++;
if (questionPhase > 2) {
throw new EncogError("Only two ?'s may be used.");
}
}
if (layer.getCount() == 0) {
throw new EncogError(
"Unknown architecture element: " + architecture + ", can't parse: " + part);
}
result.addLayer(new BasicLayer(af, bias, layer.getCount()));
}
}
result.getStructure().finalizeStructure();
result.reset();
return result;
}
public static BasicNetwork generateNetwork() {
final BasicNetwork network = new BasicNetwork();
network.addLayer(new BasicLayer(MultiBench.INPUT_COUNT));
network.addLayer(new BasicLayer(MultiBench.HIDDEN_COUNT));
network.addLayer(new BasicLayer(MultiBench.OUTPUT_COUNT));
network.getStructure().finalizeStructure();
network.reset();
return network;
}
public void recognizer(List<File> files) {
FeatureExtractor fe = new FeatureExtractor();
MLDataSet trainingSet = new BasicMLDataSet();
for (File f : files) {
// System.out.println(f.getAbsolutePath());
List<double[]> data;
try {
data = fe.fileProcessor(f);
MLData mldataIn = new BasicMLData(data.get(0));
double[] out = new double[NUM_OUT];
Integer index = new Integer(Labeler.getLabel(f));
// System.out.println(index+""+data.get(0));
out[index] = 1.;
System.out.println(out.toString());
MLData mldataout = new BasicMLData(out);
trainingSet.add(mldataIn, mldataout);
} catch (FileNotFoundException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
BasicNetwork network = new BasicNetwork();
network.addLayer(new BasicLayer(NUM_IN));
network.addLayer(new BasicLayer(new ActivationSigmoid(), true, 4 * NUM_IN));
// network.addLayer(new BasicLayer(new ActivationSigmoid(), true, 2 * NUM_IN));
network.addLayer(new BasicLayer(new ActivationSigmoid(), false, NUM_OUT));
network.getStructure().finalizeStructure();
network.reset();
// train the neural network
ResilientPropagation train = new ResilientPropagation(network, trainingSet);
System.out.println("Training Set: " + trainingSet.size());
int epoch = 1;
do {
train.iteration();
System.out.println("Epoch:" + epoch + " Error-->" + train.getError());
epoch++;
} while (train.getError() > 0.001);
train.finishTraining();
// test the neural network
System.out.println("Neural Network Results:");
for (MLDataPair pair : trainingSet) {
final MLData output = network.compute(pair.getInput());
System.out.println(
"actual-->" + Labeler.getWord(output) + ", ideal-->" + Labeler.getWord(pair.getIdeal()));
}
Encog.getInstance().shutdown();
}
/**
* Use an array to populate the memory of the neural network.
*
* @param array An array of doubles.
* @param network The network to encode.
*/
public static void arrayToNetwork(final double[] array, final BasicNetwork network) {
int index = 0;
for (final Layer layer : network.getStructure().getLayers()) {
if (layer.hasBias()) {
// process layer bias
for (int i = 0; i < layer.getNeuronCount(); i++) {
layer.setBiasWeight(i, array[index++]);
}
}
if (network.getStructure().isConnectionLimited()) {
index = NetworkCODEC.processSynapseLimited(network, layer, array, index);
} else {
index = NetworkCODEC.processSynapseFull(network, layer, array, index);
}
}
}
@Override
public MLMethod createML(int inputs, int outputs) {
BasicNetwork network = new BasicNetwork();
network.addLayer(new BasicLayer(activation, false, inputs)); // (inputs));
for (Integer layerSize : layers) network.addLayer(new BasicLayer(activation, true, layerSize));
network.addLayer(new BasicLayer(activation, true, outputs));
network.getStructure().finalizeStructure();
network.reset();
return network;
}
@Override
protected BasicNetwork createNetwork() {
BasicNetwork network = new BasicNetwork();
network.addLayer(new BasicLayer(null, true, 22));
network.addLayer(new BasicLayer(new ActivationSigmoid(), false, 30));
network.addLayer(new BasicLayer(new ActivationSigmoid(), false, 3));
network.addLayer(new BasicLayer(null, false, 15));
network.getStructure().finalizeStructure();
network.reset();
return network;
}
/**
* Process a partially connected synapse.
*
* @param network The network to process.
* @param layer The layer to process.
* @param array The array to process.
* @param index The current index.
* @return The index after this synapse has been read.
*/
private static int processSynapseLimited(
final BasicNetwork network, final Layer layer, final double[] array, final int index) {
int result = index;
// process synapses
for (final Synapse synapse : network.getStructure().getPreviousSynapses(layer)) {
if (synapse.getMatrix() != null) {
// process each weight matrix
for (int x = 0; x < synapse.getToNeuronCount(); x++) {
for (int y = 0; y < synapse.getFromNeuronCount(); y++) {
final double oldValue = synapse.getMatrix().get(y, x);
double value = array[result++];
if (Math.abs(oldValue) < network.getStructure().getConnectionLimit()) {
value = 0;
}
synapse.getMatrix().set(y, x, value);
}
}
}
}
return result;
}
@BeforeTest
public void setup() {
network = new BasicNetwork();
network.addLayer(new BasicLayer(DTrainTest.INPUT_COUNT));
network.addLayer(new BasicLayer(DTrainTest.HIDDEN_COUNT));
network.addLayer(new BasicLayer(DTrainTest.OUTPUT_COUNT));
network.getStructure().finalizeStructure();
network.reset();
weights = network.getFlat().getWeights();
training = RandomTrainingFactory.generate(1000, 10000, INPUT_COUNT, OUTPUT_COUNT, -1, 1);
}
/**
* Generate the RSOM network.
*
* @return The neural network.
*/
public BasicNetwork generate() {
final Layer input = new BasicLayer(new ActivationLinear(), false, this.inputNeurons);
final Layer output = new BasicLayer(new ActivationLinear(), false, this.outputNeurons);
int y = PatternConst.START_Y;
final BasicNetwork network = new BasicNetwork(new SOMLogic());
network.addLayer(input);
network.addLayer(output);
input.setX(PatternConst.START_X);
output.setX(PatternConst.START_X);
input.setY(y);
y += PatternConst.INC_Y;
output.setY(y);
network.getStructure().finalizeStructure();
network.reset();
return network;
}
/**
* Construct a network analyze class. Analyze the specified network.
*
* @param network The network to analyze.
*/
public AnalyzeNetwork(final BasicNetwork network) {
final int assignDisabled = 0;
final int assignedTotal = 0;
final List<Double> biasList = new ArrayList<Double>();
final List<Double> weightList = new ArrayList<Double>();
final List<Double> allList = new ArrayList<Double>();
for (int layerNumber = 0; layerNumber < network.getLayerCount() - 1; layerNumber++) {
final int fromCount = network.getLayerNeuronCount(layerNumber);
final int fromBiasCount = network.getLayerTotalNeuronCount(layerNumber);
final int toCount = network.getLayerNeuronCount(layerNumber + 1);
// weights
for (int fromNeuron = 0; fromNeuron < fromCount; fromNeuron++) {
for (int toNeuron = 0; toNeuron < toCount; toNeuron++) {
final double v = network.getWeight(layerNumber, fromNeuron, toNeuron);
weightList.add(v);
allList.add(v);
}
}
// bias
if (fromCount != fromBiasCount) {
final int biasNeuron = fromCount;
for (int toNeuron = 0; toNeuron < toCount; toNeuron++) {
final double v = network.getWeight(layerNumber, biasNeuron, toNeuron);
biasList.add(v);
allList.add(v);
}
}
}
for (final Layer layer : network.getStructure().getLayers()) {
if (layer.hasBias()) {
for (int i = 0; i < layer.getNeuronCount(); i++) {}
}
}
this.disabledConnections = assignDisabled;
this.totalConnections = assignedTotal;
this.weights = new NumericRange(weightList);
this.bias = new NumericRange(biasList);
this.weightsAndBias = new NumericRange(allList);
this.weightValues = EngineArray.listToDouble(weightList);
this.allValues = EngineArray.listToDouble(allList);
this.biasValues = EngineArray.listToDouble(biasList);
}
/**
* Process a fully connected synapse.
*
* @param network The network to process.
* @param layer The layer to process.
* @param array The array to process.
* @param index The current index.
* @return The index after this synapse has been read.
*/
private static int processSynapseFull(
final BasicNetwork network, final Layer layer, final double[] array, final int index) {
int result = index;
// process synapses
for (final Synapse synapse : network.getStructure().getPreviousSynapses(layer)) {
if (synapse.getMatrix() != null) {
// process each weight matrix
for (int x = 0; x < synapse.getToNeuronCount(); x++) {
for (int y = 0; y < synapse.getFromNeuronCount(); y++) {
synapse.getMatrix().set(y, x, array[result++]);
}
}
}
}
return result;
}
/**
* The main method.
*
* @param args No arguments are used.
*/
public static void main(final String args[]) {
// create a neural network, without using a factory
BasicNetwork network = new BasicNetwork();
network.addLayer(new BasicLayer(null, true, 2));
network.addLayer(new BasicLayer(new ActivationSigmoid(), true, 3));
network.addLayer(new BasicLayer(new ActivationSigmoid(), false, 1));
network.getStructure().finalizeStructure();
network.reset();
// create training data
MLDataSet trainingSet = new BasicMLDataSet(XOR_INPUT, XOR_IDEAL);
// train the neural network
final StochasticGradientDescent train = new StochasticGradientDescent(network, trainingSet);
train.setUpdateRule(new RMSPropUpdate());
int epoch = 1;
do {
train.iteration();
System.out.println("Epoch #" + epoch + " Error:" + train.getError());
epoch++;
} while (train.getError() > 0.01);
train.finishTraining();
// test the neural network
System.out.println("Neural Network Results:");
for (MLDataPair pair : trainingSet) {
final MLData output = network.compute(pair.getInput());
System.out.println(
pair.getInput().getData(0)
+ ","
+ pair.getInput().getData(1)
+ ", actual="
+ output.getData(0)
+ ",ideal="
+ pair.getIdeal().getData(0));
}
PerturbationFeatureImportanceCalc d;
Encog.getInstance().shutdown();
}
public static void main(String args[]) {
int inputNeurons = CHAR_WIDTH * CHAR_HEIGHT;
int outputNeurons = DIGITS.length;
BasicNetwork network = new BasicNetwork();
Layer inputLayer = new BasicLayer(new ActivationLinear(), false, inputNeurons);
Layer outputLayer = new BasicLayer(new ActivationLinear(), true, outputNeurons);
network.addLayer(inputLayer);
network.addLayer(outputLayer);
network.getStructure().finalizeStructure();
(new RangeRandomizer(-0.5, 0.5)).randomize(network);
// train it
NeuralDataSet training = generateTraining();
Train train = new TrainAdaline(network, training, 0.01);
int epoch = 1;
do {
train.iteration();
System.out.println("Epoch #" + epoch + " Error:" + train.getError());
epoch++;
} while (train.getError() > 0.01);
//
System.out.println("Error:" + network.calculateError(training));
// test it
for (int i = 0; i < DIGITS.length; i++) {
int output = network.winner(image2data(DIGITS[i]));
for (int j = 0; j < CHAR_HEIGHT; j++) {
if (j == CHAR_HEIGHT - 1) System.out.println(DIGITS[i][j] + " -> " + output);
else System.out.println(DIGITS[i][j]);
}
System.out.println();
}
}
