/// <summary>
/// Extracts features of the song. Saves it to database if the features do not exist there yet
/// </summary>
/// <param name="data">Song of which features will be extracted</param>
/// <param name="meanVector">Extracted mean vector</param>
/// <param name="mfccCoefficients">Extracted MFCC coefficients</param>
/// <param name="covarianceMatrix">Extracted covariance matrix</param>
public void Extract(DataElement data, out double[] meanVector, out double[][] mfccCoefficients, out double[,] covarianceMatrix)
{
string songName = data.FilePath.Split(new[] { '\\' }).Last();
using (var db = new Database())
{
bool songInDatabase = db.Exists(songName);
if (!songInDatabase)
{
db.AddTestSong(data.FilePath);
}
meanVector = db.GetMeanVector(songName);
covarianceMatrix = db.GetCovarianceMatrix(songName);
mfccCoefficients = db.GetMFCCCoefficients(songName);
if (!songInDatabase)
{
db.DeleteTestFiles();
}
}
}
public override SongGenre GetClass(DataElement testedData)
{
base.GetClass(testedData);
if (!testedData.HasFeaturesExtracted())
{
testedData.GetFeatures();
}
for (int i = 0; i < TrainingData.Count; i++)
{
TrainingData[i].GetDistance(testedData);
}
TrainingData.Sort((data1, data2) => data1.NeighbourDistance.CompareTo(data2.NeighbourDistance));
List <int> labelHistogram = new int[Enum.GetNames(typeof(SongGenre)).Length].ToList();
List <DataElement> nearestNeighbours = TrainingData.GetRange(0, K);
foreach (DataElement neighbour in nearestNeighbours)
{
SongGenre genre = neighbour.GetClass();
labelHistogram[(int)genre]++;
}
int max = labelHistogram.Max();
List <int> nearestGenre = new List <int>();
for (int i = 0; i < labelHistogram.Count; i++)
{
if (labelHistogram[i] == max)
{
nearestGenre.Add(i);
}
}
nearestNeighbours.RemoveAll(data => !nearestGenre.Contains((int)data.GetClass()));
Random r = new Random();
testedData.Probability = 1.0;
return(nearestNeighbours[r.Next(nearestNeighbours.Count)].GetClass());
}
private int findClosestMean(DataElement data, List <DataElement> means)
{
double minDistance = double.MaxValue;
int bestMeanIndex = -1;
int i = 0;
foreach (var m in means)
{
// zwykła odl. miedzy wektorami
//data.GetDistance(m);
//double distance = data.NeighbourDistance;
double distance = MathHelper.EuclideanDistance(data.MeanVector, m.MeanVector);
if (distance < minDistance)
{
minDistance = distance;
bestMeanIndex = i;
}
i++;
}
return(bestMeanIndex);
}
public override double GetProbability(DataElement data)
{
GetClass(data);
return(data.Probability);
}
/// <summary> /// Gets probability thad specified song belongs to classificator (important in GMM implementation) /// </summary> /// <returns>Value in range [0, 1]</returns> public abstract double GetProbability(DataElement data);
/// <summary>
/// Computes KL distance to the other song
/// </summary>
/// <param name="data">Song to which distance will be computed</param>
public void GetDistance(DataElement data)
{
NeighbourDistance = MathHelper.KLdistance(MeanVector, CovarianceMatrix, data.MeanVector, data.CovarianceMatrix);
}
/// <summary>
/// Computes genre of specified song using currently chosen classificator
/// </summary>
/// <param name="testData">Song</param>
/// <returns>Song genre converted to string</returns>
public string ComputeGenre(DataElement testData)
{
SongGenre genre = classificator.GetClass(testData);
return(DataElement.ToString(genre));
}
public override SongGenre GetClass(DataElement data)
{
throw new NotImplementedException();
//TODO:: wyznaczenie gatunku testowanego utworu
// TODO:: ustawienie data.Probability = <prawdopodobieństwo_przynależenia_do_znalezionego_gatunku>
}
public override double?GetProbability(DataElement data)
{
return(data.Probability);
}
