public void ComputeGradient()
{
Matrix.ValidateMatricesDims(NextLayer.Values.Extra, Values.Extra);
for (int raw = 0; raw < Values.Rows; raw++)
{
for (int column = 0; column < Values.Columns; column++)
{
double x = Values.Primal[raw, column]; // Current value
double dy = Values.Extra[raw, column]; // Current gradient
double df = _activation.Gradient(NextLayer.Values.Primal[raw, column], dy);
double dw = x * df;
Biases.Extra[raw, column] = df;
Weights.Extra[raw, column] = dw;
if (PrevLayer != null)
{
PrevLayer.Values.Extra[raw, column] += df * Weights.Primal[raw, column]; // Take the gradient in output unit and chain it with the local gradients . This will allow us to possibly use the output of one gate multiple times (think of it as a wire branching out), since it turns out that the gradients from these different branches just add up when computing the final gradient with respect to the circuit output.
}
}
}
#if DEBUG
if (GeneralSettings.GradientsTracingEnabled)
{
Trace.Write("Affine layer: " + GetHashCode() + ". Values grad:");
Trace.Write(Values.Extra.ToString());
Trace.Write("Bias grad: " + Biases.Extra);
Trace.WriteLine("Weights grad: " + Weights.Extra);
}
#endif
}
