public void PredictionTest()
{
var enn = new ExtendedNearestNeighbors <double[]>(k: 3, clusters: 2);
enn.Metric = (x, y) =>
{
double sumOfSquaredDifference = 0;
for (int i = 0; i < x.Length; i++)
{
sumOfSquaredDifference += Math.Pow(x[i] - y[i], 2);
}
return(Math.Sqrt(sumOfSquaredDifference));
};
double[][] trainingData;
double[][] testingData;
int[] trainingLabels;
int[] testingLabels;
int[] PredictionLabels;
// load the training data provided by the authors of [1]
using (var fs = new FileStream(@"..\..\data\ENN\TrainingData.dat", FileMode.Open))
{
trainingData = (double[][])(new BinaryFormatter().Deserialize(fs));
}
using (var fs = new FileStream(@"..\..\data\ENN\TrainingLabel.dat", FileMode.Open))
{
trainingLabels = (int[])(new BinaryFormatter().Deserialize(fs));
}
trainingLabels = trainingLabels.BijectWithNaturals();
using (var fs = new FileStream(@"..\..\data\ENN\TestingData.dat", FileMode.Open))
{
testingData = (double[][])(new BinaryFormatter().Deserialize(fs));
}
using (var fs = new FileStream(@"..\..\data\ENN\TestingLabel.dat", FileMode.Open))
{
testingLabels = (int[])(new BinaryFormatter().Deserialize(fs));
}
testingLabels = trainingLabels.BijectWithNaturals();
using (var fs = new FileStream(@"..\..\data\ENN\PredictionLabel.dat", FileMode.Open))
{
PredictionLabels = (int[])(new BinaryFormatter().Deserialize(fs));
}
PredictionLabels = PredictionLabels.BijectWithNaturals();
// Use out implementation to calculate the the distance and label maps
enn.Train(trainingData, trainingLabels);
var predictedLabels = testingData.Select(t => enn.Compute(t)).ToArray();
Assert.That(PredictionLabels.SequenceEqual(predictedLabels));
}
public void WeightedMapTest()
{
var enn = new ExtendedNearestNeighbors <double[]>(3, 2);
enn.Metric = (x, y) =>
{
double sumOfSquaredDifference = 0;
for (int i = 0; i < x.Length; i++)
{
sumOfSquaredDifference += Math.Pow(x[i] - y[i], 2);
}
return(Math.Sqrt(sumOfSquaredDifference));
};
double[][] trainingData;
int[] trainingLabels;
// load the training data provided by the authors of [1]
using (var fs = new FileStream(@"..\..\data\ENN\TrainingData.dat", FileMode.Open))
{
trainingData = (double[][])(new BinaryFormatter().Deserialize(fs));
}
using (var fs = new FileStream(@"..\..\data\ENN\TrainingLabel.dat", FileMode.Open))
{
trainingLabels = (int[])(new BinaryFormatter().Deserialize(fs));
}
trainingLabels = trainingLabels.Select(l => l == 1 ? 0 : 1).ToArray();
double[][] expectedKNNDistances;
int[][] expectedKNNLabels;
double[] expectedClassStatistics;
// load the expected outputs
using (var fs = new FileStream(@"..\..\data\ENN\weightedKNNDistance.dat", FileMode.Open))
{
expectedKNNDistances = (double[][])(new BinaryFormatter().Deserialize(fs));
}
using (var fs = new FileStream(@"..\..\data\ENN\weightedKNNLabel.dat", FileMode.Open))
{
expectedKNNLabels = (int[][])(new BinaryFormatter().Deserialize(fs));
}
using (var fs = new FileStream(@"..\..\data\ENN\TSOri.dat", FileMode.Open))
{
expectedClassStatistics = (double[])(new BinaryFormatter().Deserialize(fs));
}
// fix class labels
for (int i = 0; i < expectedKNNLabels.Length; i++)
{
for (int j = 0; j < expectedKNNLabels[i].Length; j++)
{
if (expectedKNNLabels[i][j] == 1)
{
expectedKNNLabels[i][j] = 0;
}
if (expectedKNNLabels[i][j] == 2)
{
expectedKNNLabels[i][j] = 1;
}
}
}
// Use out implementation to calculate the the distance and label maps
enn.Train(trainingData, trainingLabels);
// Get the data from the prvate fields
var WeightedKNNDistance = (double[][])enn.GetType().GetField("WeightedKNNDistance", BindingFlags.NonPublic | BindingFlags.Instance).GetValue(enn);
var WeightedKNNLabels = (int[][])enn.GetType().GetField("WeightedKNNLabels", BindingFlags.NonPublic | BindingFlags.Instance).GetValue(enn);
var ClassStatistics = (double[])enn.GetType().GetField("ClassStatistics", BindingFlags.NonPublic | BindingFlags.Instance).GetValue(enn);
// Check equality of length
Assert.That(WeightedKNNDistance.Length, Is.EqualTo(expectedKNNDistances.Length));
Assert.That(WeightedKNNLabels.Length, Is.EqualTo(expectedKNNLabels.Length));
Assert.That(ClassStatistics.Length, Is.EqualTo(expectedClassStatistics.Length));
// Check equality of content
for (int i = 0; i < WeightedKNNDistance.Length; i++)
{
Assert.That(WeightedKNNDistance[i].SequenceEqual(expectedKNNDistances[i], new DoubleEqualityComparer(1)));
}
// Check equality of content
for (int i = 0; i < WeightedKNNLabels.Length; i++)
{
Assert.That(WeightedKNNLabels[i].SequenceEqual(expectedKNNLabels[i]));
}
// Check equality of content
for (int i = 0; i < ClassStatistics.Length; i++)
{
Assert.That(ClassStatistics.SequenceEqual(expectedClassStatistics, new DoubleEqualityComparer(1E-3)));
}
}
