/// <summary>
/// Parses the .arff file output from bextract.
/// </summary>
/// <param name="arffFile">Full path to .arff file to parse.</param>
/// <returns>List of SongDataDTOs containing extracted features.</returns>
public static List <SongDataDTO> parseArff(string arffFile)
{
List <SongDataDTO> result = new List <SongDataDTO>();
string arff = File.ReadAllText(arffFile);
string[] tokens = arff.Split(new[] { "% filename " }, StringSplitOptions.None);
//Go through each of the songs (skip the first token, that has the bextract attribute comments)
for (int i = 1; i < tokens.Length; i++)
{
string songInfo = tokens[i];
SongDataDTO dto = parseSongInfo(songInfo);
result.Add(dto);
}
return(result);
}
/// <summary>
/// Classify the given songs and output a JSON string with the results.
/// </summary>
/// <remarks>
/// Results will be in the following format:
///
/// { "ClassifierResults":
/// [ { "song":
/// {
/// "title": "path\to\file\1.mp3",
/// "energy": 0.40120655758066681,
/// "positivity": 0.47041366490774172
/// }
/// },
/// { "song":
/// {
/// "title": "path\to\file\2.mp3",
/// "energy": 0.34171391252526506,
/// "positivity": 0.28074189017885803
/// }
/// },
/// ...
/// ]
/// }
/// </remarks>
///
///
/// <param name="songPaths">Paths to the songs to classify.</param>
public string Classify(string[] songPaths)
{
if (posSvm == null || energySvm == null)
{
throw new Exception("SVMs have not yet been trained!");
}
//Get features for all the songs
List <SongDataDTO> songsWithFeatures = getFeatures(songPaths);
//Classify each song
List <ClassifierResult> classifierResults = new List <ClassifierResult>();
ConsoleManager.Show();
for (int i = 0; i < songsWithFeatures.Count; i++)
{
SongDataDTO songData = songsWithFeatures[i];
ClassifierResult result = new ClassifierResult();
Song song = new Song();
//song.title = songData.getFilename(); Turns the database into .wavs
song.title = songPaths[i];
//Convert dto to double arrays (so svm can use them)
double[] posFeatures = new double[bextractPosCols.Count()];
double[] energyFeatures = new double[bextractEnergyCols.Count()];
ConvertSongDataDtoToDoubleArrays(songData, ref posFeatures, ref energyFeatures);
//Run classification
song.positivity = posSvm.Score(posFeatures);
song.energy = energySvm.Score(energyFeatures);
result.song = song;
//Fun Debugging info
Console.WriteLine("Song: {0}\n Positivity: {1}\n Energy: {2} \n\n", song.title, song.positivity, song.energy);
classifierResults.Add(result);
}
Console.WriteLine("Press Enter to continue");
Console.ReadLine();
ConsoleManager.Hide();
EmotionSpaceDTOList emotionSpaceDtoList = new EmotionSpaceDTOList();
emotionSpaceDtoList.ClassifierResults = classifierResults;
return(JsonSerializer.serializeToJson(emotionSpaceDtoList));
}
public static List <SongDataDTO> parseArff(string arffFile)
{
List <SongDataDTO> result = new List <SongDataDTO>();
string arff = File.ReadAllText(arffFile);
string[] tokens = arff.Split(new[] { "% filename " }, StringSplitOptions.None);
//Go through each of the songs (skip the first token, that has the bextract attribute comments)
for (int i = 1; i < tokens.Length; i++)
{
string token = tokens[i];
string[] lines = token.Split('\n');
string filename = lines[0].Trim(new[] { '\n', '\r' });
List <List <Double> > extractedFeatures = new List <List <Double> >();
//Line two is the sampling rate, which we do not use
//Go through each of the lines of the bextract features
for (int j = 2; j < lines.Length; j++)
{
string feature_line = lines[j];
string[] feature_strs = feature_line.Split(',');
List <Double> features = new List <Double>();
//Convert each feature into a double, add to the double array
//Skip the last one, it will be 'music'
for (int k = 0; k < feature_strs.Length - 1; k++)
{
string feature_str = feature_strs[k];
Double feature = Double.Parse(feature_str);
features.Add(feature);
}
extractedFeatures.Add(features);
}
SongDataDTO dto = new SongDataDTO(filename, extractedFeatures);
result.Add(dto);
}
return(result);
}
/// <summary>
/// Extract the data from the SongDataDTO object and place into arrays of doubles. This is needed for training.
/// </summary>
/// <remarks>
/// Only pulls out the bextract columns for positivity/energy.
/// This function will allocate the arrays.
/// </remarks>
/// <param name="songData">SongDataDTO object containing the bextract data to use for training.</param>
/// <param name="posFeatures">Reference to double array to store the features needed for training positivity.</param>
/// <param name="energyFeatures">Reference to double array to store the features needed for training energy.</param>
private void ConvertSongDataDtoToDoubleArrays(SongDataDTO songData, ref double[] posFeatures, ref double[] energyFeatures)
{
//Stick them in double arrays
List <List <Double> > features = songData.getFeatures();
List <Double> featureList = features[0]; //Just pull the first row of features for now
posFeatures = new double[bextractPosCols.Count() + 1];
energyFeatures = new double[bextractEnergyCols.Count() + 1];
//Pull out relevant bextract columns
for (int i = 1; i < bextractPosCols.Count(); i++)
{
int col = bextractPosCols.ElementAt(i);
posFeatures[i] = featureList[col];
}
for (int i = 1; i < bextractEnergyCols.Count(); i++)
{
int col = bextractEnergyCols.ElementAt(i);
energyFeatures[i] = featureList[col];
}
}
/// <summary>
/// Parses a song's section from the .arff file output of.
/// </summary>
/// <param name="songInfo">The song section of the .arff file. Between "% filename " tokens.</param>
/// <returns>SongDataDTO containing extracted features</returns>
private static SongDataDTO parseSongInfo(string songInfo)
{
string[] lines = songInfo.Split('\n');
//Line 1 is the filename
string filename = lines[0].Trim(new[] { '\n', '\r' });
//Line two is the sampling rate, which we do not use
//Following lines are extracted features for each sample of the song
List <List <Double> > extractedFeatures = new List <List <Double> >();
for (int i = 2; i < lines.Length; i++)
{
string featureLine = lines[i];
List <Double> features = parseFeatureLine(featureLine);
extractedFeatures.Add(features);
}
SongDataDTO dto = new SongDataDTO(filename, extractedFeatures);
return(dto);
}