Régis LEBRUN

Not really, as a\in[a,a]. Think about a discrete distribution.

If the test x\in[a,b] is done using x>=a and xb

As @jschueller said, the arithmetic of floats is neither associative nor commutative, so depending on the order the values are aggregated in sums, the results will differ, but in well...

The `BlockIndependentDistribution` is definitely the class to use. Nevertheless, you can use the ComposedDistribution the way you used it, but you have to tweak it a little bit to make...

At least. The specific code is efficient only when F^{-1} and c are almost always singular at 0 and 1. I was too optimistic when I implemented this algorithm: it...

@jschueller Is it normal to use a `TruncatedNormalFactory` on a sample generated by a `FisherSnedecor` for an example of use of the `FisherSnedecorFactory` ;-)? I fix it ASAP

So do I have to wait for the integration of #2072 before to rework this one?

I made the modification and synchronized the branch with the current master, and then I get the following error during the tests: ``` File "/home/regis/openturns-dev/build_improve_truncated_normal/install/lib64/python3.8/site-packages/openturns/dist_bundle2.py", line 9731, in openturns.dist_bundle2.MethodOfMomentsFactory Failed...

Thanks for the careful reading of the doc. Some complements are in order here: - From a practical point of view, there is no actual reason to stick to the...

Thanks for the link. The implementation based on FFT needs to work with data points distributed on a lattice, which involves a binning procedure: you replace your unevenly distributed points...