private void addFoldsDataToTableAndTestTable(FuzzyTable table, FuzzyTable testDataTable, ArrayList[] foldsInstances, int fold, int numberOfFolds, FuzzyTable fuzzyTable)
{
for (int i = 0; i < numberOfFolds; i++)
{
if (i != fold)
{
var foldData = foldsInstances[i];
for (int j = 0; j < foldData.Count; j++)
{
var index = (int)foldData[j];
var data = fuzzyTable.GetTable().Rows[index];
table.GetTable().ImportRow(data);
}
}
else
{
var foldData = foldsInstances[i];
for (int j = 0; j < foldData.Count; j++)
{
var index = (int)foldData[j];
var data = fuzzyTable.GetTable().Rows[index];
testDataTable.GetTable().ImportRow(data);
}
}
}
}
public void CalculateResultForRules(FuzzyTable testData, List <Rule> rules, ConfusionMatrix confusionMatrix, double tolerance = .5)
{
Classificator classificator = new Classificator();
for (int i = 0; i < testData.GetTable().Rows.Count; i++)
{
var predictedResult = classificator.Classify(testData, i, rules);
var predicterPositiveResult = predictedResult[testData.PositiveColumn.Id];
var predicterNegativeResult = predictedResult[testData.NegativeColumn.Id];
var actualPositiveResult = testData.GetPositiveColumn(i);
var actualNegativeResult = testData.GetNegativeColumn(i);
if (Math.Abs(predicterPositiveResult - actualPositiveResult) < tolerance)
{
confusionMatrix.TruePositiveCount++;
}
else
{
confusionMatrix.FalseNegativeCount++;
}
if (Math.Abs(predicterNegativeResult - actualNegativeResult) < tolerance)
{
confusionMatrix.TrueNegativeCount++;
}
else
{
confusionMatrix.FalsePositiveCount++;
}
}
}
public double[] getClassValuesNumber(FuzzyAttributeLabel[] classValues, int numberOfClassValues, FuzzyTable fuzzyTable)
{
double[] countClass = new double[numberOfClassValues];
for (int i = 0; i < classValues.Length; i++)
{
var data = this.getData(fuzzyTable, classValues[i], i);
}
for (int i = 0; i < fuzzyTable.GetTable().Rows.Count; i++)
{
var instance = fuzzyTable.GetTable().Rows[i];
for (int j = 0; j < classValues.Length; j++)
{
countClass[j] += this.getData(fuzzyTable, classValues[j], i);
// var classAttributeValue = this.getData(fuzzyTable, classValues[j], i);
// if (classValues[j].Equals( ""+((int)classAttributeValue))) {
// countClass[j] += this.getData(fuzzyTable, classValues[i], i);
// }
}
}
return(countClass);
}
public ConfusionMatrix Validate(int numberOfFolds, FuzzyTable fuzzyTable, IProcessable algorithm, double tolerance = .5)
{
int instancesSize = fuzzyTable.GetTable().Rows.Count;
ArrayList[] foldsInstances = new ArrayList[numberOfFolds];
for (int i = 0; i < numberOfFolds; i++)
{
foldsInstances[i] = new ArrayList();
}
int numberOfClassValues = fuzzyTable.getClassAttribute().Labels.Length;
FuzzyAttributeLabel[] classValues = new FuzzyAttributeLabel[numberOfClassValues];
for (int i = 0; i < numberOfClassValues; i++)
{
classValues[i] = fuzzyTable.getClassAttribute().Labels[i];
}
double[] countClass = getClassValuesNumber(classValues, numberOfClassValues, fuzzyTable);
int foldSize = instancesSize / numberOfFolds;
double[] foldClassSize = new double[countClass.Length];
for (int i = 0; i < foldClassSize.Length; i++)
{
double perc = countClass[i] / (double)instancesSize;
foldClassSize[i] = (foldSize * perc);
}
var dataCountInOneReplication = fuzzyTable.DataCount() / numberOfFolds; // the size of the fold
var confusionMatrix = new ConfusionMatrix();
var noDataTable = fuzzyTable.CloneNoData();
var rngIndexes = getRNGIndexes(instancesSize);
for (int i = 0; i < numberOfFolds; i++)
{
var instancesAdded = new ArrayList(foldSize); // what will be deleted
double[] foldClassSizeAdded = new double[foldClassSize.Length];
foreach (int index in rngIndexes)
{
var label1Value = this.getData(fuzzyTable, classValues[0], index); // e.g c1= 0.8
var label2Value = this.getData(fuzzyTable, classValues[1], index); // e.g c2= 0.2
if (label1Value > 0.5) // c1 > 0.5
{
if (foldClassSizeAdded[0] < foldClassSize[0])
{
foldClassSizeAdded[0] += label1Value;
foldsInstances[i].Add(index); // add the index to the fold
instancesAdded.Add(index);
}
}
else // c2 > 0.5
{
if (foldClassSizeAdded[1] < foldClassSize[1])
{
foldClassSizeAdded[1] += label2Value;
foldsInstances[i].Add(index); // add the index to the fold
instancesAdded.Add(index);
}
}
if (foldClassSizeAdded[0] >= foldClassSize[0] && foldClassSizeAdded[1] >= foldClassSize[1])
{
break;
}
}
// remove indexes that were used in this fold
foreach (var item in instancesAdded)
{
rngIndexes.Remove(item);
}
}
// now i have the folds
for (var fold = 0; fold < numberOfFolds; fold++)
{
var table = (FuzzyTable)fuzzyTable.CloneNoData();
var testDataTable = (FuzzyTable)table.CloneNoData();
addFoldsDataToTableAndTestTable(table, testDataTable, foldsInstances, fold, numberOfFolds, fuzzyTable);
algorithm.init(table);
var rules = algorithm.process();
if (!ExistsAtLeastOneRuleForEachClassAttribute(table, rules))
{
return(null);
}
CalculateResultForRules(testDataTable, rules, confusionMatrix, tolerance);
}
// Console.WriteLine("Accuracy: "+confusionMatrix.Accuracy());
// Console.WriteLine("Sensitivity: "+confusionMatrix.Sensitivity());
// Console.WriteLine("Specificity: "+confusionMatrix.Specificity());
// Console.WriteLine("Precision: "+confusionMatrix.Precision());
// Console.WriteLine("Krit: "+confusionMatrix.Criteria());
// Console.WriteLine("Kriteria: "+(confusionMatrix.Sensitivity() + confusionMatrix.Specificity()) / 2);
confusionMatrix.CalculatePercentNumbers();
return(confusionMatrix);
}