tiadams / java-neural-networks

Java implementations of a number of NN types

Technical Neural Networks

Build

This project is build with maven

build with mvn clean compile assembly:single

execute main class with java -jar target/multilayer-perceptron-1.0-SNAPSHOT-jar-with-dependencies.jar <TYPE_OF_DEMO (either MLP or RBF)>

Multilayer Perceptron

// define network
int[] shape = {2, 10, 10, 1};
TransferFuncType[] functions = {TransferFuncType.IDENTITY, TransferFuncType.TANH, TransferFuncType.TANH, TransferFuncType.TANH};
MLPNetwork testMLP = new MLPNetwork(shape, functions);

// load Patterns
PatternLoader loader = new PatternLoader();
List<Pattern> patterns = loader.loadPatterns(new File("src/main/resources/training.dat"));

// shuffle
List<Pattern> shuffled = loader.getShuffledPatternList(patterns_clean);

// train
testMLP.train(shuffled);

// visualize
ErrorPlotter plotter = new ErrorPlotter(testMLP.errorValues);
plotter.showErrorPlot();

Radial Basis Function Network

int[] shape2 = {2, 2, 1};
double[][] centers = {{0.0, 1.0}, {1.0, 0.0}};
RBFNetwork testRBF = new RBFNetwork(shape2, centers);
double[] testPoint = {0.0, 0.0};
System.out.println("Testing XOR for "+Arrays.toString(testPoint)+": "+Arrays.toString(testRBF.calculateOutputs(testPoint)));
double[] testPoint2 = {1.0, 0.0};
System.out.println("Testing XOR for "+Arrays.toString(testPoint2)+":"+Arrays.toString(testRBF.calculateOutputs(testPoint2)));