Michael Baudin

Technical description of DiceKriging estimation : [v51i01.pdf](https://github.com/openturns/openturns/files/3617125/v51i01.pdf)

The previous example is adapted from : http://openturns.github.io/openturns/master/examples/data_analysis/sample_correlation.html which is wrong.

Using the $\beta$ upper quantile of the binomial distribution with parameters $n$ and $\alpha$ should be much faster than using a `for` loop.

The build fails on Linux because of an error unrelated to the current PR.

The doc part is managed in PR #1484. The implementation part remains to be done.

I do not know if this is LGPL, but it claims to be faster: https://kdepy.readthedocs.io/en/latest/comparison.html

I have many doubts at the criterias used in the benchmark: * Bandwidth Selection * Available Kernels * Multidimension * Heterogeneous data * FFT-based computation * Tree-based computation The "number...

I added OT in the benchmark: https://nbviewer.jupyter.org/github/mbaudin47/KDEpy/blob/BenchmarkAddOT/docs/presentation/figs/profiling.ipynb The results are interesting:  When the binning enters, the timing stops to increase, then increases again for really large sample sizes. I...

This relates to the issue #1568, which could be considered to be solved when this PR is merged.

@efekhari27 : May you point to the otwrapy example?