Multivariate Normal/Inverse-Wishart distribution

[akonneker]

Do you have any plans to implement the distributions in the title? I'm working on implementing some things from Simon Prince's Computer Vision, and those are the only two of the common conjugate-pair distributions he uses that aren't already implemented in statrs, as far as I can tell.

[boxtown]

There were definitely plans to implement both, I just haven't had a lot of time recently. I can see if I can get one of the two into master by the end of this week and maybe the other next week. As always, I 100% welcome anyone who wishes to contribute

[akonneker]

I can take a crack at it, but I don't have much experience implementing statistical distributions. I'll look at the normal and gamma implementations and see if I can muddle my way through making them multidimensional.

…

On Tue, Oct 2, 2018 at 11:28 AM Michael Ma ***@***.***> wrote: There were definitely plans to implement both, I just haven't had a lot of time recently. I can see if I can get one of the two into master by the end of this week and maybe the other next week. As always, I 100% welcome anyone who wishes to contribute — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#90 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAR3oJAsRJZLiFBGza6WuBzq-2x95CQHks5ug5Q-gaJpZM4XDQGk> .

[boxtown]

After doing some investigating, this will take a little more work than I initially thought, mainly in that currently statrs provides no support for matrices / linear algebra. It is clear that the better choice would be to depend on one of the already existing libraries out there for working with matrices, but the question is which one to use. I'll need to take some time to properly evaluate the current landscape as I want to be very judicious about adding dependencies

[akonneker]

Thank you for looking into this. I've been pretty slammed at work, so I didn't make much progress on my end. Nalgebra seems like the most likely candidate to me re:matrix support, but that seems like a fairly hefty dependency to add, so I can understand your caution.

My impression was that things in the rust linear algebra space were somewhat in limbo until const generics dropped, but the last time I looked into it was at least a year ago. Nalgebra seems to have made nice progress since then, and the other libraries seem to have gone dormant.

[vks]

My impression was that things in the rust linear algebra space were somewhat in limbo until const generics dropped

AFAIK, const generics are still not implemented: rust-lang/rust#44580

[boxtown]

It looks like Nalgrebra seems to be the best maintained/most popular library, I'll experiment with incorporating it as a dependency in order to implement one of these distributions and report back. Hopefully it doesn't take me a month this time

[vks]

The multivariate normal distribution was implemented by now, using nalgebra. The inverse Wishart distribution still needs to be implemented.

[henryjac]

Inverse Wishart (and Wishart) was implemented in #170