Julien Schueller

ah, in that case that's not available because conda-forge has not migrated all the packages yet

@syam5g we have osx/arm binaries via conda-forge now (but still no pip), could you test it ?

no solution yet

I think we could add support for discrete distributions as long as we keep it consistent with the OT behavior: return 0 outside the support It can be added in...

@regislebrun, look at the the result of interval1.intersect(interval2): it has a>b

so `[-2.22475e-05, -40.1203]` should be considered as an empty interval, and computeprobability is actually correct to return 0 but still it's weird to get a>b

a=b works too

ok

results can differ because of ordering of computations you should still get the same results when you force 1 thread though, else it's a bug

you mean openblas threads (lapack is not parallelized), that's already taken care of, see TBBContext class