Add full covariance matrix support

[ebattenberg]

Will need to adapt sklearn full covariance routines to CUDAMat.

[ArtemiyFirsov]

Hey!

Is this repo going to be supported?

[ebattenberg]

Is this repo going to be supported?

Unfortunately, I won't be working on it, but I'd be glad to review contributions.

[ArtemiyFirsov]

@ebattenberg ok, thank you. And is are these scripts well tested?) If have some time, will work on that I guess)

[ArtemiyFirsov]

@ebattenberg BTW, when I try to estimate gmm from 100.000 samples with 26 dimensions of floats, I get CUDABlas error with no explanation. Using 1 GPU with 12GB of memory, it should be enough and ensured, that I shutdown cm sessions.

Actually, it works sometimes, not always))).

[ebattenberg]

I haven't run the code in about 4 years, so I can't guarantee that it will continue to work with current versions of CUDA. Sorry.

…

On Tue, Oct 10, 2017 at 11:24 PM Artemiy Firsov ***@***.***> wrote: @ebattenberg <https://github.com/ebattenberg> BTW, when I try to estimate gmm from 100.000 samples with 26 dimensions, I get CUDABlas error with no explanation — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABUEeuDC9xuYWNhzc4yljCVcGuDwkcCaks5srF8agaJpZM4CRSEq> .