Guillaume Dalle
Guillaume Dalle
Yes that's what I meant by "there should be a tangent on the distribution object itself, but I don't know how to do it (yet)", I'm not super at ease...
In the meantime @devmotion would you be open to adding `gradlogpdf` to the docs, or is it voluntarily absent from the public API?
Alright then, I had no idea about all of this history! I'll just draft a PR to add `gradlogpdf` to the Distributions doc if that's okay
Is there a reason not to put this one in the API though?
But I guess the cases that are implemented in Distributions correspond to probability measures with explicit log-density gradient. For those, I think it would make sense to at least mention...
@malthesr I would find this feature very useful to parallelize statistical estimation! Happy to help or review the PR if needed
Thanks for your answer. Indeed, the behavior you describe seems logical too, since sufficient statistics are computed from the sample. But I agree that both options can make sense, and...
Hum, after testing it doesn't quite go the way you said it would :sweat_smile: ```julia julia> using Distributions julia> x, w = Float32[1, 2, 3], Float32[4, 5, 6]; julia> fit(Normal{Float32},...
I think one would also need to make sufficient stats parametric, for instance here: https://github.com/JuliaStats/Distributions.jl/blob/dd6ae8f4eac304f404b0069540a6c3bb1c667f92/src/univariate/continuous/normal.jl#L108-L113
I toyed with this concept a few months ago, and I still have some leftover code for a `MultivariateCategorical`, which is almost the same. Ping me if you want it!