public static double Error(this Network network, TrainingSet[] trainingSets, double[][][] weights)
{
double error = 0;
for (var i = 0; i < trainingSets.Length; i++)
{
error += network.Error(trainingSets[i], weights);
}
return(error);
}
public static double[][][] OptimizeSingle(Network network, TrainingSet[] trainingSets, double[][][] weights, int layerIndex, int neuronIndex, int inputIndex, double delta = 10, int recursions = 10)
{
//Console.WriteLine("Optimizing: [" + layerIndex + ", " + neuronIndex + ", " + inputIndex + "]...");
var bestError = network.Error(trainingSets, weights);
for (var r = 0; r < recursions; r++)
{
//Console.WriteLine("START: " + delta + " " + weights[layerIndex][neuronIndex][inputIndex]);
if (bestError == 0)
{
break;
}
weights[layerIndex][neuronIndex][inputIndex] = weights[layerIndex][neuronIndex][inputIndex] + delta;
var positiveDeltaError = network.Error(trainingSets, weights);
weights[layerIndex][neuronIndex][inputIndex] = weights[layerIndex][neuronIndex][inputIndex] - (2.0 * delta);
var negativeDeltaError = network.Error(trainingSets, weights);
if (Math.Min(negativeDeltaError, positiveDeltaError) < bestError)
{
bestError = Math.Min(negativeDeltaError, positiveDeltaError);
if (negativeDeltaError < positiveDeltaError)
{
}
else
{
weights[layerIndex][neuronIndex][inputIndex] = weights[layerIndex][neuronIndex][inputIndex] + (2.0 * delta);
}
}
else
{
weights[layerIndex][neuronIndex][inputIndex] = weights[layerIndex][neuronIndex][inputIndex] + delta;
delta = delta * 0.5;
}
//Console.WriteLine("END: " + delta + " " + weights[layerIndex][neuronIndex][inputIndex]);
}
Console.WriteLine("Current best error: " + bestError);
return(weights);
}
public static double[][][] OptimizeSingle(Network network, TrainingSet[] trainingSets, double[][][] weights, int layerIndex, int neuronIndex, int inputIndex, double delta = 10, int recursions = 10)
{
//Console.WriteLine("Optimizing: [" + layerIndex + ", " + neuronIndex + ", " + inputIndex + "]...");
var bestError = network.Error(trainingSets, weights);
for (var r = 0; r < recursions; r++)
{
//Console.WriteLine("START: " + delta + " " + weights[layerIndex][neuronIndex][inputIndex]);
if (bestError == 0) break;
weights[layerIndex][neuronIndex][inputIndex] = weights[layerIndex][neuronIndex][inputIndex] + delta;
var positiveDeltaError = network.Error(trainingSets, weights);
weights[layerIndex][neuronIndex][inputIndex] = weights[layerIndex][neuronIndex][inputIndex] - (2.0 * delta);
var negativeDeltaError = network.Error(trainingSets, weights);
if (Math.Min(negativeDeltaError, positiveDeltaError) < bestError)
{
bestError = Math.Min(negativeDeltaError, positiveDeltaError);
if (negativeDeltaError < positiveDeltaError)
{
}
else
{
weights[layerIndex][neuronIndex][inputIndex] = weights[layerIndex][neuronIndex][inputIndex] + (2.0 * delta);
}
}
else
{
weights[layerIndex][neuronIndex][inputIndex] = weights[layerIndex][neuronIndex][inputIndex] + delta;
delta = delta * 0.5;
}
//Console.WriteLine("END: " + delta + " " + weights[layerIndex][neuronIndex][inputIndex]);
}
Console.WriteLine("Current best error: " + bestError);
return weights;
}
