001 package aima.test.learningtest.neural;
002
003 import junit.framework.TestCase;
004 import aima.learning.framework.DataSet;
005 import aima.learning.framework.DataSetFactory;
006 import aima.learning.neural.BackPropLearning;
007 import aima.learning.neural.FeedForwardNeuralNetwork;
008 import aima.learning.neural.IrisDataSetNumerizer;
009 import aima.learning.neural.IrisNNDataSet;
010 import aima.learning.neural.NNConfig;
011 import aima.learning.neural.NNDataSet;
012 import aima.learning.neural.Numerizer;
013 import aima.learning.neural.Perceptron;
014 import aima.learning.neural.Vector;
015 import aima.util.Matrix;
016
017 public class BackPropagationTests extends TestCase {
018
019 public void testFeedForwardAndBAckLoopWorks() {
020 // example 11.14 of Neural Network Design by Hagan, Demuth and Beale
021 Matrix hiddenLayerWeightMatrix = new Matrix(2, 1);
022 hiddenLayerWeightMatrix.set(0, 0, -0.27);
023 hiddenLayerWeightMatrix.set(1, 0, -0.41);
024
025 Vector hiddenLayerBiasVector = new Vector(2);
026 hiddenLayerBiasVector.setValue(0, -0.48);
027 hiddenLayerBiasVector.setValue(1, -0.13);
028
029 Vector input = new Vector(1);
030 input.setValue(0, 1);
031
032 Matrix outputLayerWeightMatrix = new Matrix(1, 2);
033 outputLayerWeightMatrix.set(0, 0, 0.09);
034 outputLayerWeightMatrix.set(0, 1, -0.17);
035
036 Vector outputLayerBiasVector = new Vector(1);
037 outputLayerBiasVector.setValue(0, 0.48);
038
039 Vector error = new Vector(1);
040 error.setValue(0, 1.261);
041
042 double learningRate = 0.1;
043 double momentumFactor = 0.0;
044 FeedForwardNeuralNetwork ffnn = new FeedForwardNeuralNetwork(
045 hiddenLayerWeightMatrix, hiddenLayerBiasVector,
046 outputLayerWeightMatrix, outputLayerBiasVector);
047 ffnn.setTrainingScheme(new BackPropLearning(learningRate,
048 momentumFactor));
049 ffnn.processInput(input);
050 ffnn.processError(error);
051
052 Matrix finalHiddenLayerWeights = ffnn.getHiddenLayerWeights();
053 assertEquals(-0.265, finalHiddenLayerWeights.get(0, 0), 0.001);
054 assertEquals(-0.419, finalHiddenLayerWeights.get(1, 0), 0.001);
055
056 Vector hiddenLayerBias = ffnn.getHiddenLayerBias();
057 assertEquals(-0.475, hiddenLayerBias.getValue(0), 0.001);
058 assertEquals(-0.1399, hiddenLayerBias.getValue(1), 0.001);
059
060 Matrix finalOutputLayerWeights = ffnn.getOutputLayerWeights();
061 assertEquals(0.171, finalOutputLayerWeights.get(0, 0), 0.001);
062 assertEquals(-0.0772, finalOutputLayerWeights.get(0, 1), 0.001);
063
064 Vector outputLayerBias = ffnn.getOutputLayerBias();
065 assertEquals(0.7322, outputLayerBias.getValue(0), 0.001);
066
067 }
068
069 public void xtestFeedForwardAndBAckLoopWorksWithMomentum() {
070 // example 11.14 of Neural Network Design by Hagan, Demuth and Beale
071 Matrix hiddenLayerWeightMatrix = new Matrix(2, 1);
072 hiddenLayerWeightMatrix.set(0, 0, -0.27);
073 hiddenLayerWeightMatrix.set(1, 0, -0.41);
074
075 Vector hiddenLayerBiasVector = new Vector(2);
076 hiddenLayerBiasVector.setValue(0, -0.48);
077 hiddenLayerBiasVector.setValue(1, -0.13);
078
079 Vector input = new Vector(1);
080 input.setValue(0, 1);
081
082 Matrix outputLayerWeightMatrix = new Matrix(1, 2);
083 outputLayerWeightMatrix.set(0, 0, 0.09);
084 outputLayerWeightMatrix.set(0, 1, -0.17);
085
086 Vector outputLayerBiasVector = new Vector(1);
087 outputLayerBiasVector.setValue(0, 0.48);
088
089 Vector error = new Vector(1);
090 error.setValue(0, 1.261);
091
092 double learningRate = 0.1;
093 double momentumFactor = 0.5;
094 FeedForwardNeuralNetwork ffnn = new FeedForwardNeuralNetwork(
095 hiddenLayerWeightMatrix, hiddenLayerBiasVector,
096 outputLayerWeightMatrix, outputLayerBiasVector);
097
098 ffnn.setTrainingScheme(new BackPropLearning(learningRate,
099 momentumFactor));
100 ffnn.processInput(input);
101 ffnn.processError(error);
102
103 Matrix finalHiddenLayerWeights = ffnn.getHiddenLayerWeights();
104 assertEquals(-0.2675, finalHiddenLayerWeights.get(0, 0), 0.001);
105 assertEquals(-0.4149, finalHiddenLayerWeights.get(1, 0), 0.001);
106
107 Vector hiddenLayerBias = ffnn.getHiddenLayerBias();
108 assertEquals(-0.4775, hiddenLayerBias.getValue(0), 0.001);
109 assertEquals(-0.1349, hiddenLayerBias.getValue(1), 0.001);
110
111 Matrix finalOutputLayerWeights = ffnn.getOutputLayerWeights();
112 assertEquals(0.1304, finalOutputLayerWeights.get(0, 0), 0.001);
113 assertEquals(-0.1235, finalOutputLayerWeights.get(0, 1), 0.001);
114
115 Vector outputLayerBias = ffnn.getOutputLayerBias();
116 assertEquals(0.6061, outputLayerBias.getValue(0), 0.001);
117
118 }
119
120 public void xtestDataSetPopulation() throws Exception {
121 DataSet irisDataSet = DataSetFactory.getIrisDataSet();
122 Numerizer numerizer = new IrisDataSetNumerizer();
123 NNDataSet innds = new IrisNNDataSet();
124
125 innds.createExamplesFromDataSet(irisDataSet, numerizer);
126
127 NNConfig config = new NNConfig();
128 config.setConfig(FeedForwardNeuralNetwork.NUMBER_OF_INPUTS, 4);
129 config.setConfig(FeedForwardNeuralNetwork.NUMBER_OF_OUTPUTS, 3);
130 config.setConfig(FeedForwardNeuralNetwork.NUMBER_OF_HIDDEN_NEURONS, 6);
131 config.setConfig(FeedForwardNeuralNetwork.LOWER_LIMIT_WEIGHTS, -2.0);
132 config.setConfig(FeedForwardNeuralNetwork.UPPER_LIMIT_WEIGHTS, 2.0);
133
134 FeedForwardNeuralNetwork ffnn = new FeedForwardNeuralNetwork(config);
135 ffnn.setTrainingScheme(new BackPropLearning(0.1, 0.9));
136
137 ffnn.trainOn(innds, 10);
138
139 innds.refreshDataset();
140 int[] result = ffnn.testOnDataSet(innds);
141 // System.out.println(result[0] + " right, " + result[1] + " wrong");
142
143 }
144
145 public void testPerceptron() throws Exception {
146 DataSet irisDataSet = DataSetFactory.getIrisDataSet();
147 Numerizer numerizer = new IrisDataSetNumerizer();
148 NNDataSet innds = new IrisNNDataSet();
149
150 innds.createExamplesFromDataSet(irisDataSet, numerizer);
151
152 Perceptron perc = new Perceptron(3, 4);
153
154 perc.trainOn(innds, 10);
155
156 innds.refreshDataset();
157 int[] result = perc.testOnDataSet(innds);
158 // System.out.println(result[0] + " right, " + result[1] + " wrong");
159
160 }
161 }