public static double ComputeOutput(double[] trainingData, double[] weight)
{
double z = 0.0;
int i = 0;
for (i = 0; i < trainingData.Length; i++)
{
z += trainingData[i] * weight[i];
}
z += weight[i];
return SGDHlper.Output(z);
}
public double[][] ComputeOutputForEachLayers(double[] trainingData)
{
var inputDataForLayers = new double[Weights.Length + 1][];
inputDataForLayers[0] = trainingData;
var count = 0;
Weights.ToList().ForEach(weights => {
inputDataForLayers[count + 1] = weights.Select(x => SGDHlper.ComputeOutput(inputDataForLayers[count], x)).ToArray();
count++;
});
return inputDataForLayers;
}
public void Train(double[] trainingData, double[] expectedResuts)
{
var newWeights = new double[Weights.Length][][];
for (int i = 0; i < newWeights.Length; i++)
{
newWeights[i] = new double[Weights[i].Length][];
for (int j = 0; j < newWeights[i].Length; j++)
{
newWeights[i][j] = new double[Weights[i][j].Length];
for (int k = 0; k < newWeights[i][j].Length; k++)
{
newWeights[i][j][k] = Weights[i][j][k];
}
}
}
var inputDataForLayers = new double[Weights.Length+1][];
var count = 0;
inputDataForLayers[0] = trainingData;
Weights.ToList().ForEach(y => {
inputDataForLayers[count+1] = y.Select(x => SGDHlper.ComputeOutput(inputDataForLayers[count], x)).ToArray();
count++;
});
var outputDelta = new double[expectedResuts.Length];
SGDHlper.TrainOutputLayer(LearningRate, inputDataForLayers[inputDataForLayers.Length-2], inputDataForLayers[inputDataForLayers.Length - 1], expectedResuts, newWeights[newWeights.Length - 1], outputDelta);
var inNodeDelta = outputDelta;
for (int i = newWeights.Length - 2; i >= 0; i--)
{
outputDelta = new double[newWeights[i].Length];
SGDHlper.TrainHiddenLayer(LearningRate, inputDataForLayers[i], inputDataForLayers[i+1], Weights[i + 1], newWeights[i], inNodeDelta, outputDelta);
inNodeDelta = outputDelta;
}
Weights= newWeights;
}