private void trainHumansToolStripMenuItem_Click_1(object sender, EventArgs e)
{
//XML file
humanList = new List <Human>();
humanList = AuxiliaryFunctions.Deserialize <List <Human> >("Database.xml");
trainArray = new double[humanList.Count][];
outputArray = new double[humanList.Count];
fileCount = humanList.Count;
InitializeProgressBar();
for (int i = 0; i < humanList.Count; i++)
{
trainArray[i] = AuxiliaryFunctions.ByteArrayToDouble(humanList[i].HOG);
outputArray[i] = humanList[i].IsHuman;
index++;
if (index % (fileCount / 100) == 0)
{
progressValue++;
label1.Text = (progressValue + " %").ToString();
pb.Value = progressValue <= 100 ? progressValue : 100;
Thread.Sleep(10);
}
}
//trainArray = humanList.Select(x => AuxiliaryFunctions.ByteArrayToDouble(x.HOG)).ToArray();
//outputArray = humanList.Select(x => (double)x.IsHuman).ToArray();
var teacher = new SequentialMinimalOptimization <Gaussian>()
{
UseComplexityHeuristic = true,
UseKernelEstimation = true
};
SupportVectorMachine <Gaussian> svm = teacher.Learn(trainArray, outputArray);
resultLine = svm.Weights;
AuxiliaryFunctions.WriteWeight(resultLine, "weight.txt");
AuxiliaryFunctions.MakeSerialization(svm, "SVM_G.xml");
}
static void Main(string[] args)
{
//Music directory
string directory = @"T:\Documents\music-classifier\clips_45seconds";
//Expected energy output
//Input file is comma separated in the following format:
// song1,expectedEnergy,expectedPositivity
// song2,expectedEnergy,expectedPositivity
//
//song1 is expected to be the filename of the song, and the file is expected to be in the same directory as the csv
//so if song1 is "hello.mp3", a hello.mp3 file should be in the directory that the csv is located
List <Song> expectedOutputs = new List <Song>();
string[] energyLines = System.IO.File.ReadAllLines(@"T:\Documents\music-classifier\clips_45seconds\expected_energy_positivity_random.csv");
foreach (string line in energyLines)
{
string[] parts = line.Split(',');
string songPath = Path.Combine(directory, parts[0]);
double expectedEnergy = double.Parse(parts[1]);
double expectedPositivity = double.Parse(parts[2]);
Classifier.Song song = new Classifier.Song();
song.positivity = expectedPositivity;
song.energy = expectedEnergy;
song.title = songPath;
expectedOutputs.Add(song);
}
//Train
Classifier.SupportVectorMachine classifier = new Classifier.SupportVectorMachine();
classifier.Train(expectedOutputs);
//Save classifier
classifier.SaveModels(Path.Combine(ExecutableInformation.getTmpPath(), "positivity.svm"), Path.Combine(ExecutableInformation.getTmpPath(), "energy.svm"));
}
static void Main(string[] args)
{
//Create and load classifier
Classifier.SupportVectorMachine svm = new Classifier.SupportVectorMachine();
svm.LoadModels(Path.Combine(@"T:\Documents\music-classifier\EmotionalGUI\resources\models", "positivity_gaussian.svm"),
Path.Combine(@"T:\Documents\music-classifier\EmotionalGUI\resources\models", "energy_gaussian.svm"));
string directory = @"T:\Documents\music-classifier\clips_45seconds";
List <Classifier.Song> expectedOutputs = new List <Classifier.Song>();
string[] songLines = System.IO.File.ReadAllLines(@"T:\Documents\music-classifier\clips_45seconds\expected_energy_positivity_random_test.csv");
double energy_total = 0;
double energy_mean = 0;
double energy_SStot = 0;
double energy_SSreg = 0;
double energy_SSres = 0;
double energy_r_squared = 0;
double energy_accuracy = 0;
double energy_rmse = 0;
int energy_correct = 0;
List <double> energy_predicted = new List <double>();
List <double> energy_stds = new List <double>();
List <double> energy_expected = new List <double>();
double positivity_total = 0;
double positivity_mean = 0;
double positivity_SStot = 0;
double positivity_SSreg = 0;
double positivity_SSres = 0;
double positivity_r_squared = 0;
double positivity_accuracy = 0;
double positivity_rmse = 0;
int positivity_correct = 0;
List <double> positivity_predicted = new List <double>();
List <double> positivity_stds = new List <double>();
List <double> positivity_expected = new List <double>();
List <string> songPaths = new List <string>();
foreach (string line in songLines)
{
//Get expected values
string[] parts = line.Split(',');
string songPath = Path.Combine(directory, parts[0]);
double expectedEnergy = double.Parse(parts[1]);
double expectedPositivity = double.Parse(parts[2]);
double energyStd = double.Parse(parts[3]);
double positivityStd = double.Parse(parts[4]);
songPaths.Add(songPath);
energy_expected.Add(expectedEnergy);
energy_stds.Add(energyStd);
positivity_expected.Add(expectedPositivity);
positivity_stds.Add(positivityStd);
}
//Classify
string jsonOutput = svm.Classify(songPaths.ToArray());
//Retrieve the results
JsonDTO jsonDTO = new System.Web.Script.Serialization.JavaScriptSerializer().Deserialize <JsonDTO>(jsonOutput);
List <Framework.ClassifierResult> classifierResults = jsonDTO.ClassifierResults;
//Get stastics
//Assuming output is in same order as input (it should be)
foreach (Framework.ClassifierResult result in classifierResults)
{
double predictedEnergy = result.song.energy;
double predictedPositivity = result.song.positivity;
energy_total += predictedEnergy;
positivity_total += predictedPositivity;
energy_predicted.Add(predictedEnergy);
positivity_predicted.Add(predictedPositivity);
}
energy_mean = energy_total / energy_predicted.Count();
positivity_mean = positivity_total / positivity_predicted.Count();
for (int i = 0; i < positivity_predicted.Count(); i++)
{
double predictedEnergy = energy_predicted[i];
double expectedEnergy = energy_expected[i];
double energyStd = energy_stds[i];
double predictedPositivity = positivity_predicted[i];
double expectedPositivity = positivity_expected[i];
double positivityStd = positivity_stds[i];
energy_correct += Math.Abs(predictedEnergy - expectedEnergy) < energyStd ? 1 : 0;
energy_SSres += Math.Pow(predictedEnergy - expectedEnergy, 2);
energy_SSreg += Math.Pow(predictedEnergy - energy_mean, 2);
energy_SStot += Math.Pow(expectedEnergy - energy_mean, 2);
positivity_correct += Math.Abs(predictedPositivity - expectedPositivity) < positivityStd ? 1 : 0;
positivity_SSres += Math.Pow(predictedPositivity - expectedPositivity, 2);
positivity_SSreg += Math.Pow(predictedPositivity - positivity_mean, 2);
positivity_SStot += Math.Pow(expectedPositivity - positivity_mean, 2);
}
energy_accuracy = (double)energy_correct / energy_predicted.Count();
energy_r_squared = 1 - energy_SSreg / energy_SStot;
energy_rmse = Math.Pow(energy_SStot / energy_predicted.Count(), 0.5);
positivity_accuracy = (double)positivity_correct / positivity_predicted.Count();
positivity_r_squared = 1 - positivity_SSreg / positivity_SStot;
positivity_rmse = Math.Pow(positivity_SStot / positivity_predicted.Count(), 0.5);
//Print
System.Console.WriteLine("Accuracy (energy)\t=\t" + energy_accuracy * 100 + "%");
System.Console.WriteLine("Accuracy (positivity)\t=\t" + positivity_accuracy * 100 + "%");
System.Console.WriteLine("Accuracy (total)\t=\t" + energy_accuracy * positivity_accuracy * 100 + "%");
System.Console.WriteLine();
System.Console.WriteLine("RMSE (energy)\t\t=\t" + energy_rmse);
System.Console.WriteLine("RMSE (positivity)\t=\t" + positivity_rmse);
System.Console.WriteLine();
System.Console.WriteLine("R^2 (energy)\t\t=\t" + energy_r_squared);
System.Console.WriteLine("R^2 (positivity)\t=\t" + positivity_r_squared);
System.Console.ReadKey();
}
