private double[] GetOutputs()
{
Layer lastLayer = layers[layers.Length - 1];
double[] outputs = new double[lastLayer.Neurons.Length];
Array.ForEach(NetMath.Range(lastLayer.Neurons.Length), i => outputs[i] = lastLayer.Neurons[i].Val);
return(outputs);
}
private double CalculateError(double[] outputs, double[] expected) //For one line of data
//Must Propagate() first
{
double[] squaredDifferenceArray = new double[outputs.Length];
for (int i = 0; i < outputs.Length; i++)
{
squaredDifferenceArray[i] = Math.Pow(expected[i] - outputs[i], 2);
}
double avgSquaredDifference = NetMath.Sum(squaredDifferenceArray) / outputs.Length;
return(avgSquaredDifference);
}
public void Propagate()
{
for (int i = 1; i < layers.Length; i++)
{
double[] prevLayerValues = new double[layers[i - 1].Neurons.Length];
Array.ForEach(NetMath.Range(layers[i - 1].Neurons.Length), j => prevLayerValues[j] = layers[i - 1].Neurons[j].Val);
for (int j = 0; j < layers[i].Neurons.Length; j++)
{
double dotSum = NetMath.DotProduct(prevLayerValues, layers[i].Neurons[j].Weights);
double activated = Sigmoid(dotSum + layers[i].Neurons[j].Bias);
layers[i].Neurons[j].Val = activated;
}
}
}
private double GetDatasetError(Dataset dataset)
{
//Calculate average error for entire dataset
double[] errors = new double[dataset.numOfDataPoints];
for (int i = 0; i < dataset.numOfDataPoints; i++)
{
double[] inputs = dataset.Inputs[i];
double[] expected = dataset.Expected[i];
double[] outputs = OutVal(inputs);
double error = CalculateError(outputs, expected);
errors[i] = error;
}
double averageError = NetMath.Sum(errors) / dataset.numOfDataPoints;
return(averageError);
}
private void SetInputs(double[] inputs)
{
Array.ForEach(NetMath.Range(inputs.Length), i => layers[0].Neurons[i].Val = inputs[i]);
}