class RooNDKeysPdf: public RooAbsPdf

 Construct N-dimensional kernel estimation p.d.f. in observables 'varList'
 from dataset 'data'. Options can be

   'a' = Use adaptive kernels (width varies with local event density)
   'm' = Mirror data points over observable boundaries. Improves modeling
         behavior at edges for distributions that are not close to zero
         at edge
   'd' = Debug flag
   'v' = Verbose flag

 The parameter rho (default = 1) provides an overall scale factor that can
 be applied to the bandwith calculated for each kernel. The nSigma parameter
 determines the size of the box that is used to search for contributing kernels
 around a given point in observable space. The nSigma parameters is used
 in case of non-adaptive bandwidths and for the 1st non-adaptive pass for
 the calculation of adaptive keys p.d.f.s.

 The optional weight arguments allows to specify an observable or function
 expression in observables that specifies the weight of each event.

 Backward compatibility constructor for (1-dim) RooKeysPdf. If you are a new user,
 please use the first constructor form.

 Backward compatibility constructor for Roo2DKeysPdf. If you are a new user,
 please use the first constructor form.

 Constructor

 evaluation order of constructor.

 set the configuration

 initialization

 copy the dataset and calculate some useful variables

 determine mirror dataset.
 mirror points are added around the physical boundaries of the dataset
 Two steps:
 1. For each entry, determine if it should be mirrored (the mirror configuration).
 2. For each mirror configuration, make the mirror points.

 determine points in +/- nSigma shell around the box determined by the variable
 ranges. These points are needed in the normalization, to determine probability
 leakage in and out of the box.

bi->nEventsBMSW=0.;
bi->nEventsBW=0.;

 sort entries, as needed for loopRange()

 loop over all closest point to x, as determined by loopRange()

 determine closest points to x, to loop over in evaluate()