vinecopulib/rvinecopulib

fitting "indep" instead of "tll" for hashed data

Closed this issue · 2 comments

tagas commented

Hi,

I have a strange issue when I try to fit a nonparametric copula with mult=0.1, it actually fits an "indep" copula. I'm attaching a script and data to reproduce the issue.

git_issue.zip

tnagler commented

Hey, thanks for the issue. The indep copula thing should be fixed in this branch: https://github.com/vinecopulib/rvinecopulib/tree/fix-bicop-select.

However, it appears that the margin fits are terrible anyway:

hist(kde1d::pkde1d(real_traj_data[, 1], vine_obj$margins[[1]]))

This should resemble a uniform distribution, but is far from it. I'll look into that as well.

tnagler commented

OK, the marginal distributions just seem to be super weird and cannot be approximated better using the interpolation approach in kde1d. I may work on this some time in the near future, but that will take time. In the meanwhile, you could fit the margins separately, use pseudo_obs(), and then fit the vine. This should make it more stable.