public static void Main()
{
/* Test of functionality of neural network */
Func <double, double> f = (t) => Cos(t);
Func <double, double> fprime = (t) => - Sin(t);
Func <double, double> F = (t) => Sin(t);
int hiddenLayers = 3;
annGauss gaussian3 = new annGauss(hiddenLayers);
int numData = 100;
double a = 0, b = 2 * PI;
vector x = new vector(numData);
vector y = new vector(numData);
for (int i = 0; i < numData; i++)
{
x[i] = a + (b - a) * i / (numData - 1);
y[i] = f(x[i]);
Write($"{x[i]} {y[i]} {fprime(x[i])} {F(x[i])}\n");
}
vector p = new vector(hiddenLayers * 3);
for (int i = 0; i < hiddenLayers; i++)
{
p[i * 3] = a + (b - a) * i / (hiddenLayers - 1);
p[i * 3 + 1] = 1;
p[i * 3 + 2] = 1;
}
gaussian3.p = p;
gaussian3.train(x, y);
/*
* See how well it predicts on unseen data
*/
for (double q = a; q < b; q += 0.05)
{
double predictfx = gaussian3.predict(q);
double predictfprimex = gaussian3.derivative(q);
double predictFx = gaussian3.antiDerivative(q);
Error.Write($"{q} {predictfx} {predictfprimex} {predictFx}\n");
}
} //Main
static void Main()
{
// First we create a set of tabulated data to feed to the network. Let's try with
// a simple sine function first, with evenly spaced points.
int m = 40;
double a = 0;
double b = 2 * PI;
vector xs = new vector(m);
vector ys = new vector(m);
for (int i = 0; i < m; i++)
{
xs[i] = a + (b - a) * i / (m - 1);
ys[i] = Sin(xs[i]);
// Print out the tabulated data to the output file
WriteLine("{0}\t{1}", xs[i], ys[i]);
}
// Let's feed the data to a neural network with 5 hidden neurons, with the activation
// being a predefined Gaussian, exp(-x^2).
int neurons = 5;
var annGauss = new annGauss(neurons);
// Next, let's train the network with the prepared data
double eps = 1e-3;
int nsteps = annGauss.training(xs, ys, eps);
StreamWriter writeOut = new StreamWriter("out.txt");
writeOut.WriteLine("The networks {0} neurons have been trained with {1} points.",
neurons, m);
writeOut.WriteLine("The network parameters were minimized to an accuracy of {0}.", eps);
writeOut.WriteLine("The training (minimization) was done in {0} steps", nsteps);
writeOut.WriteLine("The final parameters for the network are available in the Log.");
writeOut.Close();
// Now let's take 200 new data points and send it through the network and see how well
// it does. We will also try to find the derivative and the integral
int points = 200;
vector xTest = new vector(points);
vector yTest = new vector(points);
vector yDerivTest = new vector(points);
vector yIntTest = new vector(points);
// The new data is to be printed in a new block in the output file.
Write("\n\n");
for (int i = 0; i < points; i++)
{
xTest[i] = 2 * PI * i / (points - 1);
yTest[i] = annGauss.feedforward(xTest[i]);
yDerivTest[i] = annGauss.feedforwardDeriv(xTest[i]);
yIntTest[i] = annGauss.feedforwardInt(xTest[i]);
WriteLine("{0}\t{1}\t{2}\t{3}", xTest[i], yTest[i], yDerivTest[i],
yIntTest[i]);
}
// Print the network parameters to the Log
Error.WriteLine("Printing final parameters");
for (int i = 0; i < 3 * neurons; i++)
{
Error.WriteLine("{0}", annGauss.finalParams[i]);
}
}