compute sample metics before and after particles reduction

[KseniaBastrakova]

It makes sense to calculate the sample metrics before and after the distribution. We need to consider weights for each item. There is good package in R for it.
Weighted.Desc.Stat

[ax3l]

Just in case you need it, weighted binning in numpy's nd histograms as well as pandas + statsmodels might offer similar functionality in python [SO].

But especially moments are not too hard to implement quickly oneself: wiki

[sbastrakov]

@ax3l you are right, moments should be easily available or in the worst case easy to implement. And that linked R package seem to offer nothing more (so by itself it should definitely not be a reason to switch to R).

Additionally to that - and that is what is maybe more presented in R than python - we need metrics used in statistics, that fall into three categories:

• Distance between two weighted samples, at least 1D, better 3D and 6D as well (so far found only 1D Wasserstein in python iirc and used it for energies for the poster)
• Distance between two histograms
• Distance between two densities - here we can basically use any metric in functional space, but maybe some make more sense particularly for densities