public KFold(INetwork network, INetworkEvaluator evaluator)
{
_network = network;
_evaluator = evaluator;
_trainingSet = new TrainingDataSet(network.DataSet.TrainingSet.DataSet, network.LayerStructure.numberOfInputNodes);
_validationSet = new ValidationDataSet(network.DataSet.TrainingSet.DataSet, network.LayerStructure.numberOfInputNodes);
}
//for multiple-output classification network
public double EvaluateClassificationNetwork(ISubDataSet subDataSet)
{
double predictionRate = 0.0;
int numberOfDataRowsInTestSet = subDataSet.NumberOfDataRows;
int numberOfCorrectPredictions = 0;
//1. feed forward through test set
for (int i = 0; i < numberOfDataRowsInTestSet; i++)
{
var inputMatrix = subDataSet.InputMatrices[i];
var targetValues = subDataSet.OutputMatrices[i];
var feedForwardData = _network.FeedForward(inputMatrix);
int indexOfPredictedValues = feedForwardData.LayerOutputs.Count - 1;
int numberOfPredictedValues = feedForwardData.LayerOutputs[indexOfPredictedValues].GetLength(0);
var predictedValues = feedForwardData.LayerOutputs[indexOfPredictedValues];
var indexOfCorrectTargetValue = 0;
var indexOPredictedValue = 0;
var largestTargetValue = targetValues.Max();
indexOfCorrectTargetValue = targetValues.ToList().IndexOf(largestTargetValue);
var largestPredictedValue = predictedValues.Max();
indexOPredictedValue = predictedValues.ToList().IndexOf(largestPredictedValue);
//2. compare the categorical index of the predicted and target values
if (indexOPredictedValue == indexOfCorrectTargetValue)
{
numberOfCorrectPredictions++;
}
}
//3. Get the successful prediction %
predictionRate = (double)numberOfCorrectPredictions / (double)numberOfDataRowsInTestSet;
//Console.WriteLine("The Classification Network predicts the test set with " + (predictionRate * 100) + "% accuracy");
return(predictionRate);
}
public double EvaluateRegressionNetwork(double maximumAllowedError, ISubDataSet subDataSet)
{
int numberOfDataRowsInTestSet = subDataSet.NumberOfDataRows;
int numberOfCorrectPredictions = 0;
double predictionRate = 0.0;
//1. feed forward through test set
for (int i = 0; i < numberOfDataRowsInTestSet; i++)
{
var inputMatrix = subDataSet.InputMatrices[i];
var targetValues = subDataSet.OutputMatrices[i];
var feedForwardData = _network.FeedForward(inputMatrix);
int indexOfPredictedValues = feedForwardData.LayerOutputs.Count - 1;
int numberOfPredictedValues = feedForwardData.LayerOutputs[indexOfPredictedValues].GetLength(0);
var predictedValues = feedForwardData.LayerOutputs[indexOfPredictedValues];
for (int j = 0; j < numberOfPredictedValues; j++)
{
double predictedValue = predictedValues[j];
double targetValue = targetValues[j];
double differenceBetweenPredictedAndTargetValue = Math.Abs(predictedValue - targetValue);
//2. compare ||predicted - target|| <= maximum allowed deviance
if (differenceBetweenPredictedAndTargetValue <= maximumAllowedError)
{
numberOfCorrectPredictions++;
}
}
}
//3. Get the successful prediction %
predictionRate = (double)numberOfCorrectPredictions / (double)numberOfDataRowsInTestSet;
// Console.WriteLine("The Regression Network predicts the test set with " + (predictionRate * 100) + "% accuracy");
return(predictionRate);
}