Régis LEBRUN

Hi all, I made a comparison between FittingTest.Kolmogorov (OT) and lillietest (Matlab r2019b, stats toolbox), and I reimplemented a non-adaptive version of lillietest's MonteCarlo algorithm in Python. Here are the...

@EmmanuelArdillon @josephmure Now we are as wrong as Matlab ;-)! I checked the development with your script: ``` # -*- coding: utf-8 -*- """ Created on Thu Sep 3 18:23:24...

I agree with you. Maybe a reference to this thread should be kept visible as it may contain valuable information on Lilliefors&Kolmogorov tests.

It is due to the (naive) way we compute the Gaussian probability of the truncation interval. For the extreme cases we should use a log scale for the PDF, CDF...

@jschueller Almost done. Checking everything.

Th only remaining issue is with computeLogCharacteristicFunction(), which requires to compute the logarithm of the difference between two values of the Faddeeva function very close one to the other. The...

For sure ;-)!

@KieranDelamotte There is nothing wrong with the variance. The conditional variance is expected to be zero at the learning points, and it is essentially what you get here. Taking the...

You should use another local solver than TNC. From my experience this solver is fragile with kriging (maybe due to poor gradients?) and quite costly. Cobyla is a better choice,...

Beware of the fat that neither `BoolCollection` nor `IndicesCollection` are what they claim they are. The first one is a `Collection`, so it **is** a collection but not of `Bool`,...