private double CalculateFeatureProbability(NaiveBayesProbabilityTable table, int featureIndex)
{
int totalNumerator = 0;
for (int i = 0; i < ClassMappingDictionary.Count; i++)
{
totalNumerator += table[featureIndex][i].Numerator;
}
return(totalNumerator / (double)ClassProbabilityTable[0].Denominator);
}
private double CalculateFeatureLikelihood(NaiveBayesProbabilityTable table, int featureNumerator, int labelIndex)
{
int totalDenominator = 0;
for (int i = 0; i < table.Rows; i++)
{
totalDenominator += table[i][labelIndex].Numerator;
}
return(featureNumerator / (double)totalDenominator);
}
private void SetupFeatureProbabilityTables()
{
for (int i = 0; i < FeatureMappingDictionaries.Count; i++)
{
var probabilityTable = new NaiveBayesProbabilityTable();
for (int j = 0; j < FeatureMappingDictionaries[i].Count; j++)
{
probabilityTable.AddRow(ClassMappingDictionary.Count);
}
FeatureProbabilityTables.Add(probabilityTable);
}
}
private List <ProbabilityStorage> PredictProbabilities(List <object> x)
{
var featureIndices = new List <int>();
for (int i = 0; i < x.Count; i++)
{
var featureMappingDict = FeatureMappingDictionaries[i];
int index = featureMappingDict.GetValueOrDefault(x[i], -1);
featureIndices.Add(index);
}
var retPredictionProbabilites = new List <ProbabilityStorage>();
foreach (KeyValuePair <object, int> mappedLabel in ClassMappingDictionary)
{
double calculatedProbability = 1.0;
int labelIndex = mappedLabel.Value;
ProbabilityTableEntry labelEntry = ClassProbabilityTable[labelIndex];
double labelProbability = labelEntry.Numerator / (double)labelEntry.Denominator;
for (int i = 0; i < featureIndices.Count; i++)
{
if (featureIndices[i] == -1)
{
double toAdd = 1.0 / (FeatureProbabilityTables[i].Rows + 1);
double labelProbabilityPrime = (labelEntry.Numerator + toAdd) / (labelEntry.Denominator + 1);
}
else
{
int featureIndex = featureIndices[i];
NaiveBayesProbabilityTable currTable = FeatureProbabilityTables[i];
ProbabilityTableEntry entry = FeatureProbabilityTables[i][featureIndex][labelIndex];
double featureProbability = CalculateFeatureProbability(currTable, featureIndex);
double divisionResult = CalculateFeatureLikelihood(currTable, entry.Numerator, labelIndex);
if (divisionResult == 0)
{
double toAdd = 1.0 / currTable.Rows;
divisionResult = (entry.Numerator + toAdd) / (entry.Denominator + 1);
}
calculatedProbability *= (divisionResult * labelProbability) / featureProbability;
}
}
retPredictionProbabilites.Add(new ProbabilityStorage(calculatedProbability, mappedLabel.Key));
}
retPredictionProbabilites.Sort(ProbabilityStorage.Sort);
return(retPredictionProbabilites);
}
