public void UpdateNeuralNet(string csvString, string trainedNetName)
{
if (!File.Exists(outputDir + "dataset/" + trainedNetName + ".tmp"))
{
var lines = csvString.Split('\n');
var csv = from line in lines
select(line.Split(',')).ToArray();
double[][] inputs = new double[150][];
double[][] outputs = new double[150][];
string[] types = new string[150];
int rowIndex = 0;
foreach (var row in csv)
{
int colIndex = 0;
inputs[rowIndex] = new double[4];
foreach (var col in row)
{
if (colIndex < row.Length - 1)
{
inputs[rowIndex][colIndex] = double.Parse(col);
}
else
{
types[rowIndex] = col;
}
colIndex++;
}
rowIndex++;
}
string[] distinct = types.Select(o => o.Replace("\r", "")).Distinct().ToArray();
PutDistinctLevel(trainedNetName, distinct);
BackPropagation bp = new BackPropagation();
rowIndex = 0;
foreach (var row in types)
{
outputs[rowIndex] = new double[3];
outputs[rowIndex] = bp.getIndex(row.Replace("\r", ""), distinct);
rowIndex++;
}
// create neural network
ActivationNetwork network = new ActivationNetwork(new SigmoidFunction(0.43), 4, 8, 3);
BackPropagationLearning teacher = new BackPropagationLearning(network);
teacher.LearningRate = 0.2;
teacher.Momentum = 0;
// loop
double error = 1;
while (error > 0.01)
{
// run epoch of learning procedure
error = teacher.RunEpoch(inputs, outputs);
Console.WriteLine("error is: " + error);
// check error value to see if we need to stop
// ...
}
network.Save(outputDir + @"dataset/" + trainedNetName + ".tmp");
}
}
public string EvaluateData(string trainedNetName, string[] distinct, double[] data)
{
BackPropagation bp = new BackPropagation();
Network loadedNet = Network.Load(outputDir + @"dataset/" + trainedNetName + ".tmp");
double[] test = loadedNet.Compute(data);
return(bp.getResult(test, distinct));
}
