public static double CalculateFMeasureOneHot(Parameters parameters)
{
//Fill confusion matrix with zero values
var classesCount = parameters.Entities[0].Classes.Count;
var confusionMatrix = new List <List <int> >(classesCount);
for (int i = 0; i < classesCount; i++)
{
var confusionMatrixRow = new List <int>(classesCount);
for (int j = 0; j < classesCount; j++)
{
confusionMatrixRow.Add(0);
}
confusionMatrix.Add(confusionMatrixRow);
}
//Iterate over leave-one-out
for (int i = 0; i < parameters.Entities.Count; i++)
{
var queryEntityClasses = new List <double>();
switch (parameters.WindowType)
{
case WindowType.Fixed:
for (int j = 0; j < parameters.Entities.Count - 1; j++)
{
if (parameters.DistancesForEachElement[i][j].Value >= parameters.WindowWidth)
{
parameters.NeighborsCount = j;
break;
}
if (j == parameters.Entities.Count - 2)
{
parameters.NeighborsCount = j + 1;
break;
}
}
break;
case WindowType.Variable:
parameters.WindowWidth = parameters.DistancesForEachElement[i][parameters.NeighborsCount].Value;
break;
default:
break;
}
if (parameters.WindowWidth == 0)
{
for (int k = 0; k < classesCount; k++)
{
queryEntityClasses.Add(GetAverageOneHot(parameters, i, k));
}
}
else
{
var numerator = 0d;
var denominator = 0d;
for (int j = 0; j < classesCount; j++)
{
for (int k = 0; k < parameters.Entities.Count - 1; k++)
{
var kernel = CalculateKernel(
parameters.DistancesForEachElement[i][k].Value / parameters.WindowWidth, parameters.KernelFunctionType);
var classColumn = parameters.Entities[parameters.DistancesForEachElement[i][k].EntityIndex].Classes[j];
numerator += classColumn * kernel;
denominator += kernel;
}
if (denominator == 0)
{
queryEntityClasses.Add(GetAverageOneHot(parameters, i, j));
}
else
{
queryEntityClasses.Add(numerator / denominator);
}
}
}
//Add prediction to confusion matrix. If two columns have similar values choose first occurence
var predictedClass = queryEntityClasses.IndexOf(queryEntityClasses.Max());
confusionMatrix[parameters.Entities[i].ClassNumber][predictedClass] += 1;
}
var fMeasure = FMeasureService.CalculateFMeasure(confusionMatrix);
return(fMeasure);
}
public static double CalculateFMeasureNaive(Parameters parameters)
{
//Fill confusion matrix with zero values
var classesCount = parameters.Entities[0].Classes.Count;
var confusionMatrix = new List <List <int> >(classesCount);
for (int i = 0; i < classesCount; i++)
{
var confusionMatrixRow = new List <int>(classesCount);
for (int j = 0; j < classesCount; j++)
{
confusionMatrixRow.Add(0);
}
confusionMatrix.Add(confusionMatrixRow);
}
//Iterate over leave-one-out
for (int i = 0; i < parameters.Entities.Count; i++)
{
var queryEntityClassNumber = 0d;
switch (parameters.WindowType)
{
case WindowType.Fixed:
for (int j = 0; j < parameters.Entities.Count - 1; j++)
{
if (parameters.DistancesForEachElement[i][j].Value >= parameters.WindowWidth)
{
parameters.NeighborsCount = j;
break;
}
if (j == parameters.Entities.Count - 2)
{
parameters.NeighborsCount = j + 1;
break;
}
}
break;
case WindowType.Variable:
parameters.WindowWidth = parameters.DistancesForEachElement[i][parameters.NeighborsCount].Value;
break;
default:
break;
}
if (parameters.WindowWidth == 0)
{
queryEntityClassNumber = GetAverageNaive(parameters, i);
}
else
{
var numerator = 0d;
var denominator = 0d;
for (int j = 0; j < parameters.Entities.Count - 1; j++)
{
var kernel = CalculateKernel(
parameters.DistancesForEachElement[i][j].Value / parameters.WindowWidth, parameters.KernelFunctionType);
var classNumber = parameters.Entities[parameters.DistancesForEachElement[i][j].EntityIndex].ClassNumber;
numerator += classNumber * kernel;
denominator += kernel;
}
if (denominator == 0)
{
queryEntityClassNumber = GetAverageNaive(parameters, i);
}
else
{
queryEntityClassNumber = numerator / denominator;
}
}
//Add prediction to confusion matrix
confusionMatrix[parameters.Entities[i].ClassNumber][Convert.ToInt32(Math.Round(queryEntityClassNumber))] += 1;
}
var fMeasure = FMeasureService.CalculateFMeasure(confusionMatrix);
return(fMeasure);
}
