[mlpack] A potential project idea for GSOC 2021

[RISHABH GARG]

Thanks Marcus and Ryan for the feedback.
I will look into some recent publications for some better methods and see
if I can find something nice to work on.

Thanks,
Rishabh Garg

On Mon, Mar 15, 2021 at 9:13 PM Marcus Edel <marcus.edel at fu-berlin.de>
wrote:

> Personally, I would check some recent publications first and see if there
> is a
> method out there that performs better than other methods; maybe it makes
> sense to add that to mlpack as part of the project; if it uses the basic
> building
> blocks you mentioned, great, we can use that as a foundation to think about
> the interface that actually reuses some of the building blocks to build
> the more
> complex method.
>
> > On 14. Mar 2021, at 14:49, Ryan Curtin <ryan at ratml.org> wrote:
> >
> > On Sun, Mar 14, 2021 at 10:19:24PM +0530, RISHABH GARG wrote:
> >> Hello Marcus and Ryan, I did a bit of research and found a few pitfalls
> in
> >> the statsmodels library :-
> >>    1. The algorithms written in it are in-memory algorithms, so it is
> >> incapable of handling large datasets.
> >>    2. It does not have very good documentation.
> >>
> >> We can easily beat it in terms of documentation, but I am not sure about
> >> the external memory algorithms. Also, I would like to know if the
> >> algorithms implemented in mlpack are in-memory or external memory?
> >
> > All mlpack models use Armadillo, which only supports in-memory
> > computation, but the algorithms themselves are implemented in a generic
> > way, so with a little bit of work and hacking it is possible to use
> > external memory for mlpack computations (but I think nobody is really
> > doing this).
> >
> > --
> > Ryan Curtin    | "Hey, tell me the truth... are we still in the
> > ryan at ratml.org | game?" - The Chinese Waiter
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20210315/07551626/attachment.htm>