[mlpack] GSoC - 2013 Collaborative Filtering - Introduction and Initial thoughts

[Ryan Curtin]

On Fri, Apr 12, 2013 at 05:20:03PM -0700, Ajinkya @ GaTech wrote:
> I feel (as Ryan mentioned) even a CF package which can leverage the
> existing mlpack algorithms will
> be an interesting project and we can later decide if we want to make it
> available as a CF package under mlpack or not?
> It will be interesting to start with a few algorithms mentioned here
> http://docs.graphlab.org/collaborative_filtering.html. Having said that, a
> good starting point will be to build a base CF package using existing
> mlpack algorithms and analyze the speed and compare with other ML libraries
> which offer the same.

So, mlpack does at the current time have NMF which using ALS, and I
implemented SGD some time back (but I don't really like it as an
optimizer, in general).

If SVD is a useful approach, we can reimplement QuicSVD.  That used to
be implemented in mlpack ages ago but it didn't really have a
maintainer.  I am not familiar with the approach, but here is the paper:

http://fodava.gatech.edu/files/reports/FODAVA-09-11.pdf

-- 
Ryan Curtin       | "How long have you had these things on?"
ryan at igglybob.com | "Sixty-two years."