npstat::CensoredQuantileRegressionOnKDTree< Point, Numeric > Class Template Reference
Inheritance diagram for npstat::CensoredQuantileRegressionOnKDTree< Point, Numeric >:
Detailed Descriptiontemplate<class Point, class Numeric>
|
npstat::CensoredQuantileRegressionOnKDTree< Point, Numeric >::CensoredQuantileRegressionOnKDTree | ( | const KDTree< Point, Numeric > & | dataTree, |
const Functor1< double, Point > & | regressedValue, | ||
const Functor1< CensoringInfo, Point > & | censoringInfo, | ||
const QuadraticOrthoPolyND & | poly, | ||
double | quantile, | ||
double | valueLimit | ||
) |
Constructor arguments are as follows:
dataTree – the tree of data points.
regressedValue – a functor that provides the observed value for the given input point. Typically, this value will be just one of the Point coordinates not used in k-d tree construction.
censoringInfo – a functor that provides information about censoring for the goven point.
poly – the set of orthogonal polynomials used to construct the local regression surface.
quantile – the target quantile (between 0.0 and 1.0).
The "valueLimit" parameter plays the role of a far-away point which for sure lies below all uncut response values if the cut was response > cutoff or above all uncut values if the cut was response < cutoff. This limit should not be chosen extremely far away from the realistically possible response values because such a limit will contribute too much to the loss function and the loss value calculation will suffer excessively from round-off errors.
This object will not own "dataTree", "regressedValue", "censoringInfo", or "poly" objects. These objects must still exist when the CensoredQuantileRegressionOnKDTree object is in use.
|
virtual |
Method from AbsVisitor that we have to implement
Reimplemented from npstat::QuantileRegressionOnKDTree< Point, Numeric >.