/**
* train
* trains the network from given inputs and target outputs
* @param inputs, List<List<double>> a list of input Lists
* @param targets, List<List<double>> a list of corresponding target Lists
*/
public void Train(List <List <double> > inputs, List <List <double> > targets, TrainingAlgorithmFactory.TrainingAlgorithmType trainingAlgorithmType = TrainingAlgorithmFactory.TrainingAlgorithmType.HoldBestInvestigate)
{
var algorithm = TrainingAlgorithmFactory.CreateAlgoRithm(trainingAlgorithmType);
algorithm.Train(this, inputs, targets);
}
public static Network CreateNetwork(Data trainingData, int numLayers, int perLayer, TrainingAlgorithmFactory.TrainingAlgorithmType algorithm)
{
//Create hidden layers
var hidden = new List <int>();
for (var i = 0; i < numLayers; i++)
{
hidden.Add(perLayer);
}
//Create Network
var network = new Network(trainingData.DataPoints[0].Inputs.Count, hidden, trainingData.DataPoints[0].Outputs.Count);
//New network with 5 inputs, One hidden layer of 2 neurons, 1 output
//TODO: Training should be a seperate call <- this might explain the over fitting aswell
//Train the network
//network.Train(trainingData.Inputs(), trainingData.Outputs(), algorithm);
return(network);
}
public static string RunTestNetwork(Data data, Func <List <double>, List <double>, bool> successCondition, Func <List <double>, List <double> > deconvert, int numLayers, int perLayer, int epoch, int algorithmNumber, int callbackNumber, TrainingAlgorithmFactory.TrainingAlgorithmType algorithm, Func <Data, Tuple <Data, Data> > dataDelineationCallback, bool saveReport = true)
{
//Deliniate data
var dataTuple = dataDelineationCallback(data);
var trainingData = dataTuple.Item1;
var testingData = dataTuple.Item2;
//Create hidden layers
var hidden = new List <int>();
for (var i = 0; i < numLayers; i++)
{
hidden.Add(perLayer);
}
//Create Network
Network network = new Network(trainingData.DataPoints[0].Inputs.Count, hidden, trainingData.DataPoints[0].Outputs.Count);
network.MaxEpochs = epoch;
//Start a stopwatch
var stopWatch = new Stopwatch();
stopWatch.Start();
//Train the network
network.Train(trainingData.Inputs(), trainingData.Outputs(), algorithm);
Network.Save(network);
//Stop the stopwatch
stopWatch.Stop();
//Test
if (saveReport)
{
SaveReport(testingData, successCondition, deconvert, network);
}
var successes = testingData.Inputs().Select(t => network.Run(t)).Where((result, i) => successCondition(result, testingData.DataPoints[i].Outputs)).Count();
return(String.Format("{0}|{1}|{2}|{3}|{4}|{5}|{6}|{7}", network.Id, numLayers, perLayer, epoch, algorithmNumber, callbackNumber,
Math.Round((successes / (double)testingData.DataPoints.Count) * 100, 2),
(double)stopWatch.ElapsedMilliseconds / 1000));
}
