here's the closed issue. https://github.com/JuliaStats/GLM.jl/issues/80
On 6 July 2014 10:55, Florian Oswald <[email protected]> wrote: > hi guys, > sorry for not closing this thread earlier: @simonster pushed exactly the > patch suggested by Milan yesterday evening. haven't tried yet, but should > be all good. > cheers > florian > > > On 5 July 2014 23:26, Milan Bouchet-Valat <[email protected]> wrote: > >> Le samedi 05 juillet 2014 à 15:06 -0700, John Myles White a écrit : >> > Not really sure. Can you try printing out each of the columns to see >> > where the error is? Maybe the problem is with our new code for >> > printing shorthand rather than eps() when p is very small. (That’s >> > just total speculation, though.) >> Line 46 of statmodels.jl is indeed this: >> return @sprintf("<1e%2.2d", max(iceil(nextfloat(log10(pv))), -99)) >> >> And an InexactError can happen in some cases, for example: >> julia> iceil(nextfloat(log10(1e-400))) >> ERROR: InexactError() >> in iceil at float.jl:62 >> >> This is because log(1e-400) returns -Inf. A very simple fix would be to >> call iceil() after max() so that we call iceil(-99) instead of >> iceil(-Inf). I'll make a PR. >> >> >> Regards >> > >
