public string[] Classify(Vector[] sample)
{
var result = new List <string>( );
var classes = answers.Distinct( ).ToList( );
var classDists = new Dictionary <string, double>( );
foreach (var u in sample)
{
classDists.Clear( );
foreach (var c in classes)
{
classDists.Add(c, 0);
}
int i = 0;
foreach (var v in GetSampleUnion( ).OrderBy(v => Euclid.Dist(u, v.Values)))
{
classDists[v.Class] += CalcNeighbourWeight(u, v.Values, i++);
}
// To make class with max weight first
result.Add(classDists.OrderByDescending(pair => pair.Value).First( ).Key);
}
return(result.ToArray( ));
}
protected override double CalcNeighbourWeight(Vector u, Vector neighbour, int neighbourIndex)
{
if (u != cachedU)
{
sourceVec = train.OrderBy(x => Euclid.Dist(u, x)).ElementAt(hSource);
cachedU = u;
}
double h = Euclid.Dist(u, sourceVec);
return(kernelFunc(Euclid.Dist(u, neighbour) / h));
}
protected override double CalcNeighbourWeight(Vector u, Vector neighbour, int neighbourIndex)
{
if (u != cachedU)
{
indexOrder = (from quant in train.Select((x, i) => new { Values = x, Index = i })
orderby Euclid.Dist(u, quant.Values)
select quant.Index).ToArray( );
cachedU = u;
}
var dist = Euclid.Dist(u, neighbour) / widths[indexOrder[neighbourIndex]];
return(charges[indexOrder[neighbourIndex]] * PotentialFunction(dist));
}
protected override double CalcNeighbourWeight(Vector u, Vector neighbour, int neighbourIndex)
{
return(kernelFunc(Euclid.Dist(u, neighbour) / h));
}
public static Func <Vector, Vector, int, double> DistanceWeight( )
{
// Because we maximize weight of class, nearest object must have the biggest weight
return((u, v, i) => 1.0 / Euclid.Dist(u, v));
}
