The smoothing parameters is selected by the method of Geenens et al. (2017).
It uses principal components for the rotation matrix and selects the
nearest neighbor fraction along each principal direction by approximate
least-squares cross-validation.

bw_tll_nn(udata, deg)

## Arguments

udata |
data. |

deg |
degree of the polynomial. |

## Value

A list with entries:

`B`

rotation matrix,

`alpha`

nearest neighbor fraction (this one is multiplied
with `mult`

in `kdecop()`

),

`kappa`

correction factor for differences in roughness along
the axes,

see Geenens et al. (2017).

## References

Geenens, G., Charpentier, A., and Paindaveine, D. (2017).
Probit transformation for nonparametric kernel estimation of the copula
density.
Bernoulli, 23(3), 1848-1873.