public static void logOuts(List <double[]> ins, List <double[]> outs, NN.neuralNet myNN)
{
for (int i = 0; i < ins.Count; i++)
{
double[] output = myNN.forward(ins[i]);
for (int j = 0; j < output.Length; j++)
{
Console.WriteLine(output.arrayToString() + "," + outs[i].arrayToString());
}
}
}
public static void logFail(List <double[]> ins, List <double[]> outs, NN.neuralNet myNN)
{
for (int i = 0; i < ins.Count; i++)
{
double[] output = myNN.forward(ins[i]);
for (int j = 0; j < output.Length; j++)
{
//if (((int)Math.Round(output[j],1)) != (int)Math.Round(outs[i][j])) { Console.WriteLine(output.arrayToString() + "," + outs[i].arrayToString()); return; }
//if (((int)Math.Round(output[j])) != (int)Math.Round(outs[i][j])) { Console.WriteLine(output.arrayToString() + "," + outs[i].arrayToString()); return; }
if (!binaryCompare(output[j], outs[i][j], .08))
{
Console.WriteLine(output.arrayToString() + "," + outs[i].arrayToString()); return;
}
}
}
}
public static bool good(List <double[]> ins, List <double[]> outs, NN.neuralNet myNN)
{
for (int i = 0; i < ins.Count; i++)
{
double[] output = myNN.forward(ins[i]);
//Console.WriteLine(output.arrayToString()+","+outs[i].arrayToString());
for (int j = 0; j < output.Length; j++)
{
//Console.WriteLine(((int)Math.Round(output[j])) + "," + (int)Math.Round(outs[i][j]));
//if ((((int)Math.Round(output[j], 1))) != ((int)Math.Round(outs[i][j]))) return false;
//if ((((int)Math.Round(output[j]))) != ((int)Math.Round(outs[i][j]))) return false;
if (!binaryCompare(output[j], outs[i][j], .08))
{
return(false);
}
}
}
return(true);
}
public static void Main(string[] args)
{
//List<double> testData = new List<double> { 1, -1, 0, 1, -1, 1 };
//Console.WriteLine(testData.normalize(-1,1).ToArray().arrayToString());
//;
//Maybe neuron delta derivative can be calc by input.sum and take deriviative?
//Make training data equal, too many 0s
NN.neuralNet mynn = new NN.neuralNet();
mynn.addLayer(2, 2, 2, NN.activationFunction.input); //Input
//mynn.addLayer(2, 2, 2, NN.activationFunction.tanh, true); //Hidden
mynn.addLayer(2, 2, 2, NN.activationFunction.tanh); //Hidden
//mynn.addLayer(2, 2, 2, NN.activationFunction.tanh); //Hidden
mynn.addLayer(1, 2, 1, NN.activationFunction.tanh); //, true); //Output
//mynn.addLayer(2, 2, 3, NN.activationFunction.input); //Input
//mynn.addLayer(3, 2, 3, NN.activationFunction.sigmoid);
//mynn.addLayer(3, 3, 3, NN.activationFunction.sigmoid);
//mynn.addLayer(1, 3, 1, NN.activationFunction.sigmoid);
mynn.setLearn(NN.learningMode.staticLearn, .5);
List <double[]> ins = new List <double[]>();
List <double[]> outs = new List <double[]>();
//ins.Add(new double[] { 0, 0 }); outs.Add(new double[] { 0 });
//ins.Add(new double[] { 0, 1 }); outs.Add(new double[] { 1 });
//ins.Add(new double[] { 1, 0 }); outs.Add(new double[] { 1 });
//ins.Add(new double[] { 1, 1 }); outs.Add(new double[] { 1 });
//ins.Add(new double[] { 0, 0 }); outs.Add(new double[] { 0 });
//ins.Add(new double[] { 0, 1 }); outs.Add(new double[] { 0 });
//ins.Add(new double[] { 1, 0 }); outs.Add(new double[] { 0 });
//ins.Add(new double[] { 1, 1 }); outs.Add(new double[] { 1 });
ins.Add(new double[] { 0, 0 }); outs.Add(new double[] { 0 });
ins.Add(new double[] { 0, 1 }); outs.Add(new double[] { 1 });
ins.Add(new double[] { 1, 0 }); outs.Add(new double[] { 1 });
ins.Add(new double[] { 1, 1 }); outs.Add(new double[] { 0 });
//Console.WriteLine(new double[] { 6.5,5,3 }.ToList<double>().normalize().ToArray().arrayToString());
//return;
int c = 0;
while (true)
{
c++;
if (c >= 5000)
{
c = 0;
mynn.backward(ins, outs, false);
double error = mynn.getError(ins, outs);
Console.WriteLine("Total Error: " + error);
Console.WriteLine("Net Momentum: " + mynn.getNetMomentum());
logFail(ins, outs, mynn);
}
if (good(ins, outs, mynn))
{
Console.WriteLine("Converged!");
break;
}
}
Console.WriteLine("Weight Matrix: ");
for (int i = 1; i < mynn.layers.Count; i++)
{
Console.Write("[{0}]", i);
for (int j = 0; j < mynn.layers[i].neurons.Length; j++)
{
Console.Write(mynn.layers[i].neurons[j].weights.arrayToString() + ",");
}
Console.WriteLine();
}
Console.WriteLine("{Network Out}, {Expected Out}: ");
logOuts(ins, outs, mynn);
//if (sendEmail) mailDone("Converged");
//mynn.save("nn");
Console.WriteLine(mynn.toMath());
//Console.WriteLine(mynn.solve(new double[] { 0 }, new double[] { 0 }, mynn.getError(ins, outs), 0, 1));
//Console.ReadLine();
}