private void doNearestKNeighboursAlgorithmTest()
{
var metric = new EuclideanMetric();
var alg = new NearestKNeighboursAlgorithm(Data.TrainingSample, metric, 1);
// LOO
alg.Train_LOO();
var optK = alg.K;
Console.WriteLine("Nearest K Neigbour: optimal k is {0}", optK);
Console.WriteLine();
// Margins
Console.WriteLine("Margins:");
calculateMargin(alg);
Console.WriteLine();
//Error distribution
Console.WriteLine("Errors:");
for (int k = 1; k < 5; k++)
{
alg.K = k;
var errors = alg.GetErrors(Data.Data);
var ec = errors.Count();
var dc = Data.Data.Count;
var pct = Math.Round(100.0F * ec / dc, 2);
Console.WriteLine("{0}:\t{1} of {2}\t({3}%) {4}", k, ec, dc, pct, k == optK ? "<-LOO optimal" : string.Empty);
}
Console.WriteLine();
Visualizer.Run(alg);
}
/// <summary>
/// Leave-one-out optimization
/// </summary>
public static void OptimizeLOO(NearestKNeighboursAlgorithm alg, int?minK = null, int?maxK = null)
{
if (!minK.HasValue || minK.Value < 1)
{
minK = 1;
}
if (!maxK.HasValue || maxK.Value > alg.TrainingSample.Count)
{
maxK = alg.TrainingSample.Count - 1;
}
var kOpt = int.MaxValue;
var minErrCnt = int.MaxValue;
for (int k = minK.Value; k <= maxK.Value; k++)
{
var errCnt = 0;
alg.K = k;
var initSample = alg.TrainingSample;
foreach (var pData in initSample)
{
var looSample = initSample.ApplyMask((p, c, idx) => p != pData.Key);
try
{
alg.TrainingSample = looSample;
var predClass = alg.Predict(pData.Key);
var realClass = pData.Value;
if (!predClass.Equals(realClass))
{
errCnt++;
}
}
finally
{
alg.TrainingSample = initSample;
}
}
if (errCnt < minErrCnt)
{
minErrCnt = errCnt;
kOpt = k;
}
}
alg.K = kOpt;
}
