[mlpack] Regarding Participation in GSoC'19

[Ryan Curtin]

On Thu, Mar 14, 2019 at 10:11:41PM +0530, Subham Barnwal wrote:
> I would love to build some "Bayesian Methods Based Libraries" which i guess
> except TensorFlow probability people don't have too much options.Some of
> the things I am planning is to implement BlackBox VI,GP with
> scalability,EM(I see GMM is already there but may be I can try coding a
> more general purpose EM) etc.I am open to any other suggestions (Both
> bayesian and non-bayesian).I would appreciate your guidance on the
> same.Please help me over the same.

Hi Subham,

Thanks for getting in touch and the ideas that you've proposed look
interesting to me.  If you prepare a proposal for these things, probably
one of the most important things to focus on is the API that is provided
to users for these techniques.

Specifically for a general-purpose EM implementation, it would be great
if we could use that in place of the existing EM implementation used for
GMMs.  You can take a look at the `EMFit<>` class in
src/mlpack/methods/gmm/em_fit.hpp to see how it's currently structured.

Code reuse is a big priority for mlpack, so it would be good to take a
look through the library and see what components you could reuse or
adapt for your implementations. :)

Hope this is helpful!  Let me know if I can clarify anything.

Thanks,

Ryan

-- 
Ryan Curtin    | "So, it's just you 57 punks against KUNG FU JOE?"
ryan at ratml.org |   - Kung Fu Joe