public GaussianNaiveBayes Estimate(Matrix input, Vector output)
{
// Algorithm used to calculate variance + mean : http://i.stanford.edu/pub/cstr/reports/cs/tr/79/773/CS-TR-79-773.pdf
var distinctOutput = output.Distinct();
var nbFeatures = input.NbColumns;
var nbClasses = distinctOutput.Count();
var classCount = new int[nbClasses];
var classPrior = new int[nbClasses];
var epsilon = 1e-9 * input.Variance().Max().Key.GetNumber();
Theta = Matrix.BuildEmptyMatrix(nbClasses, nbFeatures);
Sigma = Matrix.BuildEmptyMatrix(nbClasses, nbFeatures);
for (int i = 0; i < nbClasses; i++)
{
var ot = distinctOutput.Values[i];
var indexes = output.FindIndexes(ot);
Matrix subMatrix = indexes.Select(_ => input.GetRowVector(_).Values).ToArray();
var kvp = UpdateMeanVariance(subMatrix);
Theta.SetRow(kvp.Key, i);
Sigma.SetRow(kvp.Value, i);
classCount[i] = subMatrix.NbRows;
}
Sigma = Sigma.Sum(epsilon);
CalculateClassPrior(output);
return(this);
}
