public static double[] Evaluate(Math::Matrix <double> predictions, Math::Matrix <double> shouldBe, params ErrorCalculator[] ecs)
{
List <string> diffStrings = new List <string>();
for (int row = 0; row < predictions.RowCount; row++)
{
double desiredSum = 0;
double actualSum = 0;
for (int col = 0; col < predictions.ColumnCount; col++)
{
desiredSum += shouldBe[row, col];
actualSum += predictions[row, col];
}
diffStrings.Add(String.Format("{0} {1}", desiredSum, actualSum));
}
File.WriteAllLines(@"D:\! Egyetem\! RWTH\Semester 2\Seminar CV\Datasets\Mall\analysis\differences.txt", diffStrings);
double[] results = new double[ecs.Length];
int i = 0;
foreach (ErrorCalculator ec in ecs)
{
results[i++] = Evaluate(predictions, shouldBe, ec);
}
return(results);
}
private Math::Matrix <double> LoadOrCalculateInputMatrix(IList <string> frameImageFilePaths)
{
if (CachePath == null)
{
return(CalculateInputMatrix(frameImageFilePaths));
}
string filename = String.Format("count={0}, {1}.dat", frameImageFilePaths.Count, featureExtractor.ToString());
string path = Path.Combine(CachePath, filename);
if (File.Exists(path))
{
lock (cacheLock)
{
using (var stream = File.Open(path, FileMode.Open))
{
BinaryFormatter bf = new BinaryFormatter();
return(DenseMatrix.OfArray((double[, ])bf.Deserialize(stream)));
}
}
}
else
{
lock (cacheLock)
{
Math::Matrix <double> inputMatrix = CalculateInputMatrix(frameImageFilePaths);
using (var stream = File.Open(path, FileMode.Create))
{
BinaryFormatter bf = new BinaryFormatter();
bf.Serialize(stream, inputMatrix.ToArray());
}
return(inputMatrix);
}
}
}
public Math::Matrix <double> Predict(IList <string> frameImageFilePaths)
{
Math::Matrix <double> X = LoadOrCalculateInputMatrix(frameImageFilePaths);
dataNormalizer.Normalize(X);
return(innerRegression.Predict(X));
}
public void Train(IList <string> frameImageFilePaths, IList <IList <PointF> > peoplePositions)
{
Math::Matrix <double> X = LoadOrCalculateInputMatrix(frameImageFilePaths);
dataNormalizer.TrainNormalization(X);
dataNormalizer.Normalize(X);
Math::Matrix <double> Y = PeoplePositions.GridQuantize(peoplePositions, featureExtractor.N, featureExtractor.M, 640, 480);
innerRegression.Train(X, Y);
}
public static double Evaluate(Math::Matrix <double> predictions, Math::Matrix <double> shouldBe, ErrorCalculator ec)
{
double errorSum = 0;
for (int row = 0; row < predictions.RowCount; row++)
{
double desiredSum = 0;
double actualSum = 0;
for (int col = 0; col < predictions.ColumnCount; col++)
{
desiredSum += shouldBe[row, col];
actualSum += predictions[row, col];
}
errorSum += ec.CalculateError(actualSum, desiredSum);
}
return(errorSum / (predictions.RowCount));
}
