public void movieRecommendationTestCRbm(int iterations, double learningRate, int visNodes, int hidNodes)
{
Debug.WriteLine($"{iterations}-{visNodes}-{hidNodes}");
LearningApi api = new LearningApi(getDescriptorForRbm(3898));
// Initialize data provider
api.UseCsvDataProvider(Path.Combine(Directory.GetCurrentDirectory(), @"RestrictedBolzmannMachine2\Data\movieDatasetTrain.csv"), ',', false, 0);
api.UseDefaultDataMapper();
double[] featureVector = new double[] { 0, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.8, 0.85 };
api.UseCRbm(featureVector, learningRate, iterations, visNodes, hidNodes);
Stopwatch watch = new Stopwatch();
watch.Start();
RbmScore score = api.Run() as RbmScore;
watch.Stop();
var hiddenNodes = score.HiddenValues;
var hiddenWeight = score.HiddenBisases;
double[] learnedFeatures = new double[hidNodes];
double[] hiddenWeights = new double[hidNodes];
for (int i = 0; i < hidNodes; i++)
{
learnedFeatures[i] = hiddenNodes[i];
hiddenWeights[i] = hiddenWeight[i];
}
StreamWriter tw = new StreamWriter($"PredictedDigit_I{iterations}_V{visNodes}_H{hidNodes}_learnedbias.txt");
foreach (var item in score.HiddenBisases)
{
tw.WriteLine(item);
}
tw.Close();
var testData = ReadData(Path.Combine(Directory.GetCurrentDirectory(), @"RestrictedBolzmannMachine2\Data\movieDatasetTest.csv"));
var result = api.Algorithm.Predict(testData, api.Context);
var predictedData = ((RbmResult)result).VisibleNodesPredictions;
var predictedHiddenNodes = ((RbmResult)result).HiddenNodesPredictions;
var acc = testData.GetHammingDistance(predictedData);
WriteDeepResult(iterations, new int[] { visNodes, hidNodes }, learningRate, acc, watch.ElapsedMilliseconds * 1000, predictedHiddenNodes);
WriteOutputMatrix(iterations, new int[] { visNodes, hidNodes }, learningRate, predictedData, testData);
}
public void SimpleCRBMTest()
{
LearningApi api = new LearningApi();
api.UseActionModule <object, double[][]>((notUsed, ctx) =>
{
const int maxSamples = 12;
ctx.DataDescriptor = getDescriptorForRbm_sample1();
double[][] data = new double[maxSamples][];
data[0] = new double[] { 1, 1, 0, 0, 0, 0 }; // A
data[1] = new double[] { 0, 0, 1, 1, 0, 0 }; // B
data[2] = new double[] { 0, 0, 0, 0, 1, 1 }; // C
data[3] = new double[] { 1, 1, 0, 0, 0, 1 }; // noisy A
data[4] = new double[] { 0, 0, 1, 1, 0, 0 }; // BRt
data[5] = new double[] { 0, 0, 0, 0, 1, 1 }; // C
data[6] = new double[] { 1, 0, 0, 0, 0, 0 }; // weak A
data[7] = new double[] { 0, 0, 1, 0, 0, 0 }; // weak B
data[8] = new double[] { 0, 0, 0, 0, 1, 0 }; // weak C
data[9] = new double[] { 1, 1, 0, 1, 0, 0 }; // noisy A
data[10] = new double[] { 1, 0, 1, 1, 0, 0 }; // noisy B
data[11] = new double[] { 0, 0, 1, 0, 1, 1 }; // noisy C
return(data);
});
double[] featureVector = new double[] { 1, 0 };
api.UseCRbm(featureVector, 0.01, 1000, 6, 3);
RbmScore score = api.Run() as RbmScore;
double[][] testData = new double[4][];
Assert.True(score.Loss < 1.0);
testData[0] = new double[] { 1, 1, 0, 0, 0, 0 };
testData[1] = new double[] { 0, 0, 0, 0, 1, 1 };
testData[2] = new double[] { 0, 1, 0, 0, 0, 0 };
testData[3] = new double[] { 0, 0, 0, 0, 1, 0 };
var result = api.Algorithm.Predict(testData, api.Context);
// NOT FINISHED.
//Assert.True(result[0] == 1);
//Assert.True(result[1] == 0);
//Assert.True(result[2] == 0);
//Assert.True(result[3] == 0);
//Assert.True(result[4] == 1);
//Assert.True(result[5] == 0);
}
public void smileyTestCRbm(int iterations, double learningRate, int visNodes, int hidNodes)
{
LearningApi api = new LearningApi(getDescriptorForRbm(1600));
// Initialize data provider
api.UseCsvDataProvider(Path.Combine(Directory.GetCurrentDirectory(), @"RestrictedBolzmannMachine2\Data\Smiley.csv"), ',', false, 0);
api.UseDefaultDataMapper();
double[] featureVector = new double[] { 0.1, 0.2 };
api.UseCRbm(featureVector, learningRate, iterations, visNodes, hidNodes);
Stopwatch watch = new Stopwatch();
watch.Start();
RbmScore score = api.Run() as RbmScore;
watch.Stop();
var hiddenNodes = score.HiddenValues;
var hiddenWeight = score.HiddenBisases;
double[] learnedFeatures = new double[hidNodes];
double[] hiddenWeights = new double[hidNodes];
for (int i = 0; i < hidNodes; i++)
{
learnedFeatures[i] = hiddenNodes[i];
hiddenWeights[i] = hiddenWeight[i];
}
var testData = ReadData(Path.Combine(Directory.GetCurrentDirectory(), @"RestrictedBolzmannMachine2\Data\SmileyTest.csv"));
var result = api.Algorithm.Predict(testData, api.Context);
var predictedData = ((RbmResult)result).VisibleNodesPredictions;
var predictedHiddenNodes = ((RbmResult)result).HiddenNodesPredictions;
var acc = testData.GetHammingDistance(predictedData);
var ValTest = calcDelta(predictedData, testData);
var lossTest = ValTest / (visNodes);
Debug.WriteLine($"lossTest: {lossTest}");
WriteDeepResult(iterations, new int[] { visNodes, hidNodes }, acc, watch.ElapsedMilliseconds * 1000, predictedHiddenNodes);
WriteOutputMatrix(iterations, new int[] { visNodes, hidNodes }, predictedData, testData);
}
public void CRbm_ClassifierTest()
{
var dataPath = System.IO.Path.Combine(Directory.GetCurrentDirectory(), @"RestrictedBolzmannMachine2\Data\rbm_twoclass_sample.csv");
LearningApi api = new LearningApi(this.getDescriptorForRbmTwoClassesClassifier());
// Initialize data provider
api.UseCsvDataProvider(dataPath, ';', false, 1);
api.UseDefaultDataMapper();
double[] featureVector = new double[] { 0.1, 0.2, 0.3 };
api.UseCRbm(featureVector, 0.01, 1000, 10, 2);
RbmResult score = api.Run() as RbmResult;
double[][] testData = new double[5][];
//
// This test data contains two patterns. One is grouped at left and one at almost right.
testData[0] = new double[] { 1, 1, 1, 0, 0, 0, 0, 0, 0, 0 };
testData[1] = new double[] { 1, 0, 1, 0, 0, 0, 0, 0, 0, 0 };
testData[2] = new double[] { 0, 0, 0, 0, 0, 1, 1, 1, 0, 0 };
testData[3] = new double[] { 0, 0, 0, 0, 0, 1, 0, 1, 0, 0 };
// This will be classified as third class.
testData[4] = new double[] { 1, 1, 1, 0, 0, 1, 1, 1, 0, 0 };
var result = api.Algorithm.Predict(testData, api.Context) as RbmResult;
//
// 2 * BIT1 + BIT2 of [0] and [1] should be same.
// We don't know how RBM will classiffy data. We only expect that
// same or similar pattern of data will be assigned to the same class.
// Note, we have here two classes (two hiddne nodes).
// First and second data sample are of same class. Third and fourth are also of same class.
// Here we check first classs.
Assert.True(2 * result.HiddenNodesPredictions[0][0] + result.HiddenNodesPredictions[0][1] ==
2 * result.HiddenNodesPredictions[1][0] + result.HiddenNodesPredictions[1][1]);
// Here is test for second class.
Assert.True(2 * result.HiddenNodesPredictions[2][0] + result.HiddenNodesPredictions[2][1] ==
2 * result.HiddenNodesPredictions[3][0] + result.HiddenNodesPredictions[3][1]);
printVector("Weights", result.Weights);
}