npstat::BandwidthCVPseudoLogli1D< Numeric, Num2, Num3, Num4 > Class Template Reference
Inheritance diagram for npstat::BandwidthCVPseudoLogli1D< Numeric, Num2, Num3, Num4 >:
Detailed Descriptiontemplate<typename Numeric, typename Num2, typename Num3 = double, typename Num4 = double>
|
|
inlineexplicit |
Parameter "regularizationPower" is used to limit the contributions into the overall pseudo log likelihood from points for which the "leaving one out" density is very low. For those points, instead of "leaving one out" density, we will use the density generated by that point itself divided by pow(N, regularizationPower). This method limits the effect of tails on bandwidth determination.
|
virtual |
It should be assumed that the "nFillsInRange" method of the argument histogram returns the actual number of fills (that is, the histogram represents an actual collection of points, has possible bin values of 0, 1, 2, ..., and it is not scaled).
"densityEstimate" is allowed to be an estimate without truncation (even if it includes negative values). Dependence of the optimized quantity on the bandwidth should be smoother without truncation. Naturally, the order of values in "densityEstimate" is the same as the order of bins in "histo".
The "filterUsed" parameter is needed in order to be able to do "leave-one-out" type of cross-validation. In such calculations one has to know how much of the density estimate at coordinate x is contributed by the source point at x. When density estimates are done on the grid, the filters differ for the center of the grid and on the border. Because of this, we will need to have an access to the complete filter collection.
Implements npstat::AbsBandwidthCV1D< Numeric, Num2, double, double >.
|
virtual |
Cross-validation for samples of univariate weighted points
Implements npstat::AbsBandwidthCV1D< Numeric, Num2, double, double >.