//---------MAIN LOOP-------------------//
public void run(int iterations, PerfGraph graph, bool useGraph)
{
for (int i = 0; i < weights.Length; i++)
{
initialWeights[i] = weights[i];
}
for (int i = 0; i < nodes.Count; i++)
{
for (int j = 0; j < nodes[i].weights.Length; j++)
{
initialNodeWeights[i, j] = nodes[i].weights[j];
}
}
for (int i = 0; i < iterations; i++)
{
bool done = true;
graph.updateLabel_individual("MLP on epoch:" + i);
for (int j = 0; (j < inputs.Count); j++)
{
setNodeOutputs(j);
setOutput(j);
setOutputError(j);
setNodeErrors(j);
changeOutputWeights(j);
changeInputWeights(j);
if (useGraph)
{
graph.updateRates(inputs[j].expectedOutput, inputs[j].output);
//Console.WriteLine(inputs[j].error);
}
if (Math.Abs(inputs[j].error) > 0.01)
{
done = false;
}
}
if (useGraph)
{
graph.updateMSE(meanSquaredError());
}
if (done)
{
break;
}
}
}
//---------MAIN LOOP-------------------//
public void run(int iterations, PerfGraph graph, bool useGraph)
{
for (int i = 0; i < weights.Length; i++)
{
initialWeights[i] = weights[i];
}
for(int i = 0; i < nodes.Count; i++)
{
for(int j = 0; j< nodes[i].weights.Length; j++)
{
initialNodeWeights[i, j] = nodes[i].weights[j];
}
}
for(int i = 0; i < iterations; i ++)
{
bool done = true;
graph.updateLabel_individual("MLP on epoch:" + i);
for (int j = 0; (j<inputs.Count); j++)
{
setNodeOutputs(j);
setOutput(j);
setOutputError(j);
setNodeErrors(j);
changeOutputWeights(j);
changeInputWeights(j);
if (useGraph)
{
graph.updateRates(inputs[j].expectedOutput, inputs[j].output);
//Console.WriteLine(inputs[j].error);
}
if (Math.Abs(inputs[j].error) > 0.01)
{
done = false;
}
}
if (useGraph)
{
graph.updateMSE(meanSquaredError());
}
if (done)
{
break;
}
}
}
public void run()
{
int iteration = 0;
while (iteration < iterations)
{
graph.updateLabel_evolutionary("EA on iteration " + iteration);
double[] fitness = new double[mlps.Count];
//double[,] outputs =
int i = 0;
foreach (MLP mlp in mlps)
{
graph.updateLabel_evolutionary("EA on iteration " + iteration + " training MLP:" + (i + 1));
mlp.run(50, graph, useGraph);
fitness[i] = mlp.meanSquaredError();
mlp.resetMLP();
i++;
}
bool converged = true;
double minSquaredError = 9999;
foreach (double error in fitness)
{
if (error < minSquaredError)
{
minSquaredError = error;
}
if (error != fitness[0])
{
converged = false;
}
}
graph.updateLabel_evolutionary("EA on iteration " + iteration + " min MSE:" + minSquaredError);
if (converged)
{
graph.updateLabel_evolutionary("CONVERGED after :" + iteration + " iterations with MSE:" + minSquaredError);
graph.updateLabel_individual("");
break;
}
if (evolvePopSize > 0)
{
//get the top and bottom population
List <KeyValuePair <double, MLP> > evolvingMLP = new List <KeyValuePair <double, MLP> >();
List <KeyValuePair <double, MLP> > removingMLP = new List <KeyValuePair <double, MLP> >();
for (i = 0; i < evolvePopSize; i++)
{
evolvingMLP.Add(new KeyValuePair <double, MLP>(fitness[i], mlps[i]));
removingMLP.Add(new KeyValuePair <double, MLP>(fitness[i], mlps[i]));
}
for (i = evolvePopSize; i < popSize; i++)
{
var maxBestFitness = evolvingMLP.MaxBy(kvp => kvp.Key).Key;
var maxBestFitnessPair = evolvingMLP.MaxBy(kvp => kvp.Key);
var minWorstFitness = removingMLP.MinBy(kpv => kpv.Key).Key;
var minWorstFitnessPair = removingMLP.MinBy(kpv => kpv.Key);
if (fitness[i] < maxBestFitness)
{
evolvingMLP.Remove(maxBestFitnessPair);
evolvingMLP.Add(new KeyValuePair <double, MLP>(fitness[i], mlps[i]));
}
if (fitness[i] > minWorstFitness)
{
removingMLP.Remove(minWorstFitnessPair);
removingMLP.Add(new KeyValuePair <double, MLP>(fitness[i], mlps[i]));
}
}
//evolve the top population
List <MLP> temp = new List <MLP>();
foreach (KeyValuePair <double, MLP> kvp in removingMLP)
{
temp.Add(kvp.Value);
}
//remove the population with the bottom fitness
foreach (MLP m in temp)
{
mlps.Remove(m);
}
temp.Clear();
foreach (KeyValuePair <double, MLP> kvp in evolvingMLP)
{
temp.Add(DeepClone(kvp.Value));
}
//add the evolved mlps to the population
List <MLP> evolved = evolvePop(temp);
foreach (MLP m in evolved)
{
mlps.Add(m);
}
}
iteration++;
}
}