[mlpack] Comparative analysis of novel clustering algorithms

[Ryan Curtin]

On Mon, Mar 11, 2019 at 09:12:27PM +0530, Heet Sankesara wrote:
> Hello community, Myself Heet sankesara. I am a machine learning
> practitioner. I've been doing it for a year. I am currently pursuing my
> BTech in CSE from IIIT Vadodara. I am doing research on  Markov Logic
> Network. Due to this, I have to leave python and start working with cpp.
> I've been learning mlkit for a few weeks now and have a fair idea of the
> underlying structure.
>  In GSoC ideas list, there is a clustering method Quantum clustering. I
> want to work on a few novel clustering algorithms like *sampling clustering
> <https://arxiv.org/pdf/1806.08245.pdf>*, *Deep Clustering for Unsupervised
> Learning of Visual Features <https://arxiv.org/pdf/1807.05520.pdf>*, Learning
> Neural Models for End-to-End Clustering
> <https://arxiv.org/pdf/1807.04001.pdf>, *GMM **clustering, *and *Quantum
> clustering *etc and do the comparative analysis of them. This comparative
> analysis will be aimed at knowing the strength and weakness of each
> algorithm and what kind of data is good for which algorithm. Please
> consider this idea for GSoC 2K19. I am happy to talk further and discuss
> possible algorithms which can be implemented in the upcoming summer.
> With Regards,
> Heet Sankesara

Hi Heet,

Sorry for the slow response.  I do think it would definitely be nice to
have some additional clustering algorithms in mlpack.  And it is really
important to have some comparative analysis to know what is better.  If
you put together a proposal for GSoC this year for this, I do think the
comparative analysis will be a nice thing to add, since after the summer
is over, presenting this analysis in a blog post (or something similar)
can be really helpful for mlpack users in deciding which algorithm they
want to use.

You might also consider comparing against the existing baselines we have
(regular GMM and k-means for instance).  There's also mean shift and
DBSCAN that are currently implemented.

Thanks!

Ryan

-- 
Ryan Curtin    | "Open the pig!"
ryan at ratml.org |   - Frank Moses