/// <summary>
/// Get the best estimated y for the current value.
/// </summary>
public double Y()
{
if (_model == null)
{
Estimate();
}
return(_model.Y(_x[SelfIndex]));
}
private void Estimate()
{
for (int iter = 0; iter < LoessConfiguration.T; iter++)
{
_model = Regression();
// calculate the errors
var errors = new double[NeighborsX.Length];
var absErrors = new double[NeighborsX.Length];
for (int i = 0; i < NeighborsX.Length; i++)
{
double error = NeighborsY[i] - _model.Y(NeighborsX[i]);
errors[i] = error;
absErrors[i] = Math.Abs(error);
}
Array.Sort(absErrors);
double median = absErrors[absErrors.Length / 2];
if (median == 0) // a very subtle bug! sometimes, when the input data is very clean, so that the median could be 0!
{
median = double.Epsilon;
}
// calculate the gain for new weights. the outliers will get much less weight
var deltas = new double[errors.Length];
for (int i = 0; i < errors.Length; i++)
{
deltas[i] = WeightMethod.BisquareWeight(errors[i] / 6.0 / median);
}
// update new weights.
for (int i = 0; i < Weights.Length; i++)
{
Weights[i] *= deltas[i];
}
}
}
