[1] | 1 | package geniusweb.blingbling.Ranknet4j;
|
---|
| 2 |
|
---|
| 3 | import java.util.ArrayList;
|
---|
| 4 | import java.util.List;
|
---|
| 5 |
|
---|
| 6 | import org.neuroph.core.Layer;
|
---|
| 7 | import org.neuroph.core.NeuralNetwork;
|
---|
| 8 | import org.neuroph.core.input.WeightedSum;
|
---|
| 9 | import org.neuroph.core.transfer.Linear;
|
---|
| 10 | import org.neuroph.nnet.comp.neuron.BiasNeuron;
|
---|
| 11 | import org.neuroph.nnet.comp.neuron.InputNeuron;
|
---|
| 12 | //import org.neuroph.nnet.learning.BackPropagation;
|
---|
| 13 | //import org.neuroph.nnet.learning.MomentumBackpropagation;
|
---|
| 14 | import org.neuroph.util.ConnectionFactory;
|
---|
| 15 | import org.neuroph.util.LayerFactory;
|
---|
| 16 | import org.neuroph.util.NeuralNetworkFactory;
|
---|
| 17 | import org.neuroph.util.NeuralNetworkType;
|
---|
| 18 | import org.neuroph.util.NeuronProperties;
|
---|
| 19 | import org.neuroph.util.TransferFunctionType;
|
---|
| 20 | import org.neuroph.util.random.RangeRandomizer;
|
---|
| 21 |
|
---|
| 22 | /**
|
---|
| 23 | * Multi Layer Perceptron neural network with Back propagation learning algorithm.
|
---|
| 24 | *
|
---|
| 25 | * @author Zoran Sevarac <sevarac@gmail.com>
|
---|
| 26 | */
|
---|
| 27 | public class Ranknet extends NeuralNetwork<BackPropagation> {
|
---|
| 28 |
|
---|
| 29 | /**
|
---|
| 30 | * The class fingerprint that is set to indicate serialization
|
---|
| 31 | * compatibility with a previous version of the class.
|
---|
| 32 | */
|
---|
| 33 | private static final long serialVersionUID = 2L;
|
---|
| 34 |
|
---|
| 35 | /**
|
---|
| 36 | * Creates new MultiLayerPerceptron with specified number of neurons in layers
|
---|
| 37 | *
|
---|
| 38 | * @param neuronsInLayers collection of neuron number in layers
|
---|
| 39 | */
|
---|
| 40 | public Ranknet(List<Integer> neuronsInLayers) {
|
---|
| 41 | // init neuron settings
|
---|
| 42 | NeuronProperties neuronProperties = new NeuronProperties();
|
---|
| 43 | neuronProperties.setProperty("useBias", true);
|
---|
| 44 | neuronProperties.setProperty("transferFunction", TransferFunctionType.SIGMOID);
|
---|
| 45 |
|
---|
| 46 | this.createNetwork(neuronsInLayers, neuronProperties);
|
---|
| 47 | }
|
---|
| 48 |
|
---|
| 49 | public Ranknet(int... neuronsInLayers) {
|
---|
| 50 | // init neuron settings
|
---|
| 51 | NeuronProperties neuronProperties = new NeuronProperties();
|
---|
| 52 | neuronProperties.setProperty("useBias", true);
|
---|
| 53 | neuronProperties.setProperty("transferFunction", TransferFunctionType.SIGMOID);
|
---|
| 54 | neuronProperties.setProperty("inputFunction", WeightedSum.class);
|
---|
| 55 |
|
---|
| 56 | List<Integer> neuronsInLayersVector = new ArrayList<>();
|
---|
| 57 | for (int i = 0; i < neuronsInLayers.length; i++) {
|
---|
| 58 | neuronsInLayersVector.add(Integer.valueOf(neuronsInLayers[i]));
|
---|
| 59 | }
|
---|
| 60 |
|
---|
| 61 | this.createNetwork(neuronsInLayersVector, neuronProperties);
|
---|
| 62 | }
|
---|
| 63 |
|
---|
| 64 | public Ranknet(TransferFunctionType transferFunctionType, int... neuronsInLayers) {
|
---|
| 65 | // init neuron settings
|
---|
| 66 | NeuronProperties neuronProperties = new NeuronProperties();
|
---|
| 67 | neuronProperties.setProperty("useBias", true);
|
---|
| 68 | neuronProperties.setProperty("transferFunction", transferFunctionType);
|
---|
| 69 | neuronProperties.setProperty("inputFunction", WeightedSum.class);
|
---|
| 70 |
|
---|
| 71 |
|
---|
| 72 | List<Integer> neuronsInLayersVector = new ArrayList<>();
|
---|
| 73 | for (int i = 0; i < neuronsInLayers.length; i++) {
|
---|
| 74 | neuronsInLayersVector.add(Integer.valueOf(neuronsInLayers[i]));
|
---|
| 75 | }
|
---|
| 76 |
|
---|
| 77 | this.createNetwork(neuronsInLayersVector, neuronProperties);
|
---|
| 78 | }
|
---|
| 79 |
|
---|
| 80 | public Ranknet(List<Integer> neuronsInLayers, TransferFunctionType transferFunctionType) {
|
---|
| 81 | // init neuron settings
|
---|
| 82 | NeuronProperties neuronProperties = new NeuronProperties();
|
---|
| 83 | neuronProperties.setProperty("useBias", true);
|
---|
| 84 | neuronProperties.setProperty("transferFunction", transferFunctionType);
|
---|
| 85 |
|
---|
| 86 | this.createNetwork(neuronsInLayers, neuronProperties);
|
---|
| 87 | }
|
---|
| 88 |
|
---|
| 89 | /**
|
---|
| 90 | * Creates new MultiLayerPerceptron net with specified number neurons in
|
---|
| 91 | * getLayersIterator
|
---|
| 92 | *
|
---|
| 93 | * @param neuronsInLayers collection of neuron numbers in layers
|
---|
| 94 | * @param neuronProperties neuron properties
|
---|
| 95 | */
|
---|
| 96 | public Ranknet(List<Integer> neuronsInLayers, NeuronProperties neuronProperties) {
|
---|
| 97 | this.createNetwork(neuronsInLayers, neuronProperties);
|
---|
| 98 | }
|
---|
| 99 |
|
---|
| 100 | /**
|
---|
| 101 | * Creates MultiLayerPerceptron Network architecture - fully connected
|
---|
| 102 | * feed forward with specified number of neurons in each layer
|
---|
| 103 | *
|
---|
| 104 | * @param neuronsInLayers collection of neuron numbers in getLayersIterator
|
---|
| 105 | * @param neuronProperties neuron properties
|
---|
| 106 | */
|
---|
| 107 | private void createNetwork(List<Integer> neuronsInLayers, NeuronProperties neuronProperties) {
|
---|
| 108 |
|
---|
| 109 | // set network type
|
---|
| 110 | this.setNetworkType(NeuralNetworkType.MULTI_LAYER_PERCEPTRON);
|
---|
| 111 |
|
---|
| 112 | // create input layer
|
---|
| 113 | NeuronProperties inputNeuronProperties = new NeuronProperties(InputNeuron.class, Linear.class);
|
---|
| 114 | Layer layer = LayerFactory.createLayer(neuronsInLayers.get(0), inputNeuronProperties);
|
---|
| 115 |
|
---|
| 116 | boolean useBias = true; // use bias neurons by default
|
---|
| 117 | if (neuronProperties.hasProperty("useBias")) {
|
---|
| 118 | useBias = (Boolean) neuronProperties.getProperty("useBias");
|
---|
| 119 | }
|
---|
| 120 |
|
---|
| 121 | if (useBias) {
|
---|
| 122 | layer.addNeuron(new BiasNeuron());
|
---|
| 123 | }
|
---|
| 124 |
|
---|
| 125 | this.addLayer(layer);
|
---|
| 126 |
|
---|
| 127 | // create layers
|
---|
| 128 | Layer prevLayer = layer;
|
---|
| 129 |
|
---|
| 130 | //for(Integer neuronsNum : neuronsInLayers)
|
---|
| 131 | for (int layerIdx = 1; layerIdx < neuronsInLayers.size(); layerIdx++) {
|
---|
| 132 | Integer neuronsNum = neuronsInLayers.get(layerIdx);
|
---|
| 133 | // createLayer layer
|
---|
| 134 | layer = LayerFactory.createLayer(neuronsNum, neuronProperties);
|
---|
| 135 |
|
---|
| 136 | if (useBias && (layerIdx < (neuronsInLayers.size() - 1))) {
|
---|
| 137 | layer.addNeuron(new BiasNeuron());
|
---|
| 138 | }
|
---|
| 139 |
|
---|
| 140 | // add created layer to network
|
---|
| 141 | this.addLayer(layer);
|
---|
| 142 | // createLayer full connectivity between previous and this layer
|
---|
| 143 | if (prevLayer != null) {
|
---|
| 144 | ConnectionFactory.fullConnect(prevLayer, layer);
|
---|
| 145 | }
|
---|
| 146 |
|
---|
| 147 | prevLayer = layer;
|
---|
| 148 | }
|
---|
| 149 |
|
---|
| 150 | // set input and output cells for network
|
---|
| 151 | NeuralNetworkFactory.setDefaultIO(this);
|
---|
| 152 |
|
---|
| 153 | // set learnng rule
|
---|
| 154 | // this.setLearningRule(new BackPropagation());
|
---|
| 155 | BackPropagation bp = new BackPropagation();
|
---|
| 156 | bp.setMaxIterations(4000);
|
---|
| 157 | bp.setLearningRate(0.003);
|
---|
| 158 | // bp.setMaxError(0.1);
|
---|
| 159 | // bp.setBatchMode(true);
|
---|
| 160 | this.setLearningRule(bp);
|
---|
| 161 | // this.setLearningRule(new DynamicBackPropagation());
|
---|
| 162 |
|
---|
| 163 | this.randomizeWeights(new RangeRandomizer(-1.0, 1.0));
|
---|
| 164 |
|
---|
| 165 | }
|
---|
| 166 |
|
---|
| 167 | public void connectInputsToOutputs() {
|
---|
| 168 | // connect first and last layer
|
---|
| 169 | ConnectionFactory.fullConnect(getLayerAt(0), getLayerAt(getLayersCount() - 1), false);
|
---|
| 170 | }
|
---|
| 171 |
|
---|
| 172 | } |
---|