public static TrainingDataMatlabDto GetTrainingSetMatlab(NeuralNetworkTrainingConfig trainingConfig)
{
TrainingDataMatlabDto trainingSet = new TrainingDataMatlabDto();
string line;
// Read the file and display it line by line.
System.IO.StreamReader file =
new System.IO.StreamReader(System.IO.Path.Combine(trainingConfig.TrainingDatabaseFileRoute, trainingConfig.TrainingDatabaseFileName));
while ((line = file.ReadLine()) != null)
{
var lines = line.Split(" ", StringSplitOptions.RemoveEmptyEntries);
trainingSet.XEntries.Add(Array.ConvertAll(lines.Take(trainingConfig.InputSize).ToArray(),
System.Convert.ToDouble)
.ToList());
trainingSet.YExpected.Add(Array.ConvertAll(lines.Skip(trainingConfig.InputSize)
.Take(trainingConfig.PredictedObjects.Count).ToArray(),
System.Convert.ToDouble)
.ToList());
}
file.Close();
return(trainingSet);
}
public IActionResult TrainNetworkMatlab(Guid neuralNetworkId)
{
if (!_mlpRepository.NeuralNetworkExists(neuralNetworkId))
{
return(NotFound());
}
var NeuralNetworkFromRepo = _mlpRepository.GetFullNeuralNetwork(neuralNetworkId);
TrainingDataMatlabDto trainingSet = TrainingSet.GetTrainingSetMatlab(NeuralNetworkFromRepo.TrainingConfig);
double[,] x = new double[trainingSet.XEntries.Count, NeuralNetworkFromRepo.TrainingConfig.InputSize];
double[,] d = new double[trainingSet.YExpected.Count, NeuralNetworkFromRepo.TrainingConfig.OutputNeuronElements];
for (int i = 0; i < trainingSet.XEntries.Count; i++)
{
for (int j = 0; j < NeuralNetworkFromRepo.TrainingConfig.InputSize; j++)
{
x[i, j] = trainingSet.XEntries[i][j];
}
for (int k = 0; k < NeuralNetworkFromRepo.TrainingConfig.OutputNeuronElements; k++)
{
d[i, k] = trainingSet.YExpected[i][k];
}
}
MWArray X = new MWNumericArray(x);
MWArray D = new MWNumericArray(d);
MWArray No = new MWNumericArray(NeuralNetworkFromRepo.TrainingConfig.HiddenNeuronElements);
MWArray Ns = new MWNumericArray(NeuralNetworkFromRepo.TrainingConfig.OutputNeuronElements);
MWArray eta = new MWNumericArray(NeuralNetworkFromRepo.TrainingConfig.Eta);
MWArray m = new MWNumericArray(NeuralNetworkFromRepo.TrainingConfig.Epochs);
MLP mlp = new MLP();
MWArray[] result = mlp.MLPUltra(2, X, D, No, Ns, eta, m);
//class1.untitled(a,b);
IList <NeuronForManipulation> HiddenLayer =
AutoMapper.Mapper.Map <IList <NeuronForManipulation> >(
NeuralNetworkFromRepo.Neurons
.Where(a => a.NeuronType == NeuronType.Hidden)
.OrderBy(a => a.Index));
IList <NeuronForManipulation> OutputLayer =
AutoMapper.Mapper.Map <IList <NeuronForManipulation> >(
NeuralNetworkFromRepo.Neurons
.Where(a => a.NeuronType == NeuronType.Output)
.OrderBy(a => a.Index));
foreach (NeuronForManipulation hiddenNeuron in HiddenLayer)
{
hiddenNeuron.Weights = hiddenNeuron.Weights.OrderBy(a => a.Index).ToList();
for (int i = 1; i < NeuralNetworkFromRepo.TrainingConfig.InputSize + 1; i++)
{
hiddenNeuron.Weights[i - 1].Weight = Convert.ToDouble(result[0][hiddenNeuron.Index, i + 1].ToString());
}
hiddenNeuron.Bias = Convert.ToDouble(result[0][hiddenNeuron.Index, 1].ToString());
}
foreach (NeuronForManipulation outputNeuron in OutputLayer)
{
outputNeuron.Weights = outputNeuron.Weights.OrderBy(a => a.Index).ToList();
for (int i = 1; i < NeuralNetworkFromRepo.TrainingConfig.HiddenNeuronElements + 1; i++)
{
outputNeuron.Weights[i - 1].Weight = Convert.ToDouble(result[1][outputNeuron.Index, i + 1].ToString());
}
outputNeuron.Bias = Convert.ToDouble(result[1][outputNeuron.Index, 1].ToString());
}
NeuralNetworkFromRepo.Neurons = AutoMapper.Mapper.Map <IList <Neuron> >(HiddenLayer.Concat(OutputLayer));
_mlpRepository.UpdateNeuralNetwork(NeuralNetworkFromRepo);
if (!_mlpRepository.Save())
{
throw new Exception($"Updating neural network {neuralNetworkId} failed on save.");
}
return(Ok());
}
