public void TrainModel()
{
double w1 = 0.1;
double w2 = 0.1;
double bias = 0.0;
// Conduct gradient descent
for (int j = 0; j < numInput; ++j)
{
double yHat = DLHelper.CalcLinReg(inputX1[j], inputX2[j], w1, w2, bias);
double diff = yHat - labelY[j];
// update w1, w2, and bias
w1 += gradientDescentW(inputX1[j], diff);
w2 += gradientDescentW(inputX2[j], diff);
bias += gradientDescentB(diff);
}
trainedW1 = w1;
trainedW2 = w2;
trainedBias = bias;
Console.WriteLine(string.Format("The trained W1 is {0:f4}", trainedW1));
Console.WriteLine(string.Format("The trained W2 is {0:f4}", trainedW2));
Console.WriteLine(string.Format("The trained bias is {0:f4}", trainedBias));
Console.WriteLine();
}
static void TestLinearRegression()
{
// Generate training data set.
double w1 = 2.0;
double w2 = -3.4;
double bias = 4.2;
Console.WriteLine(string.Format("The true W1 is {0:f4}", w1));
Console.WriteLine(string.Format("The true W2 is {0:f4}", w2));
Console.WriteLine(string.Format("The true bias is {0:f4}", bias));
Console.WriteLine();
int numberOfExample = 1000;
DataGenerator dg = new DataGenerator(w1, w2, bias, numberOfExample);
dg.GenerateInputAndLabel();
//string filePath = @"F:\Hanjin.Hu\Research\DeepLearning\DeepLearning\LinReg.csv";
//dg.OutputData(filePath);
List <double> inputX1 = dg.GeneratedX1.ToList();
List <double> inputX2 = dg.GeneratedX2.ToList();
List <double> labelY = dg.GeneratedLabelY.ToList();
// Train the neural network.
LinearRegressionNN lrnn = new LinearRegressionNN(inputX1, inputX2, labelY);
lrnn.TrainModel();
// Generate testing data set.
dg.Clear();
dg.GenerateInputAndLabel();
List <double> testX1 = dg.GeneratedX1.ToList();
List <double> testX2 = dg.GeneratedX2.ToList();
List <double> testY = dg.GeneratedLabelY.ToList();
// Test the model
for (int i = 0; i < 5; ++i)
{
double predictedY = lrnn.Predict(testX1[i], testX2[i]);
double linReg = DLHelper.CalcLinReg(testX1[i], testX2[i], w1, w2, bias);
//Console.WriteLine(string.Format("The real house price is {0:f6}", testY[i]));
Console.WriteLine(string.Format("The linear reg price is {0:f4}", linReg));
Console.WriteLine(string.Format("The predicted house price is {0:f4}", predictedY));
Console.WriteLine();
}
}
protected void GenerateLabelY()
{
if (generatedX1.Count == 0)
{
GenerateX1();
}
if (generatedX2.Count == 0)
{
GenerateX2();
}
double mean = 0.0;
double stdDev = 0.01;
Normal normalDist = new Normal(mean, stdDev);
for (int i = 0; i < numExamples; ++i)
{
double y = DLHelper.CalcLinReg(generatedX1[i], generatedX2[i], trueW1, trueW2, trueBias) + normalDist.Sample();
generatedLabelY.Add(y);
}
}
