Marek Kaluba

> Another thing, do you have experience with Weisberg-Bingham version? As a matter of fact, I'm not a statistician, doing pure math, coding in julia "after hours" ;-). But for...

Ok, variances/covariances turned out harder than I thought: I implemented the exact formulas from [Godwin Some Low Moments of Order Statistics](https://projecteuclid.org/euclid.aoms/1177730036) in [normordstats.jl](https://gist.github.com/kalmarek/67b6459589823dba0cacbb1aec17be8d), please have a look. These become numerically...

It took more time than I had predicted, but turned out a fun project ;-) I'm looking for a piece of advise what to do with the code, based on...

I also had too much to do, couldn't finish this cleanly;) long story short we either: 1. use Royston approximation (already in place), forgetting that he fit using 32-bits of...

@Ph0non this effort (it was fun!) is an offspring of my frustration on the teaching I was forced to do. and yes, that included teaching statistics ;-) that semester is...

There are three things to be finished here: 1. fix problems with accuracy 2. find out what Royston fit his approximation to 3. produce test suite 4. write documentation I...

@albz could You point me to a python implementation? the one from scipy https://github.com/scipy/scipy/blob/v1.1.0/scipy/stats/morestats.py#L1304 uses standard [fortran code](https://github.com/scipy/scipy/blob/v1.1.0/scipy/stats/statlib/swilk.f), the approximating polynomials were not fit in double precision there. right now...

I see that people are generally interested so here is some details. In 1992 paper Royston 1. obtained (by 10000-MC samples) what he calls "exact" coefficients (`A`), 2. took Weisberg-Bingham...

@andreasnoack Thanks for the encouragement ;) However unlikely it is, SW test is still in the back of my head. Hopefully we will be soon bringing [Arblib.jl](https://github.com/kalmarek/Arblib.jl) to a usable...

@andreasnoack in https://github.com/kalmarek/ShapiroWilk/pull/1 `SWCoeffs` are implemented using Arblib; one can do (an embarrassingly parallel) computation e.g. for `(n=50, prec=128)`, which gives in a couple of minutes ~ 7 digits of...