Xapian devel - Jan 2017 - GSoC 2017

Hello devs,

My name is Richhiey Thomas and I'm studying Computer Engineering under
Mumbai University. I had worked with Xapian in GSoC 2016 where I had worked
on Clustering of Search Results. I would want to continue working on the
project and was wondering whether it would fit the scope of GSoC.

The clustering branch had a clustering API and KMeans clusterer implemented
but hasnt been merged yet because it had to be optimized further and due to
other smaller issues. I would like to complete work on merging this
clustering branch and implementing a hierarchial clusterer.

Also, a main reason for the performance reduction with large document
corpus was because of the dimensionality of the document vectors. Therefore
a latent semantic analysis to reduce document vectors size is something
that could be necessary.

I would like to have your feedback on the same.

Thanks :)
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.xapian.org/pipermail/xapian-devel/attachments/20170123/00e3c718/attachment.html>

On 23 Jan 2017, at 12:23, Richhiey Thomas <richhiey.thomas at gmail.com>
wrote:
> My name is Richhiey Thomas and I'm studying Computer Engineering under
Mumbai University. I had worked with Xapian in GSoC 2016 where I had worked on
Clustering of Search Results. I would want to continue working on the project
and was wondering whether it would fit the scope of GSoC.
I think this will make sense as another GSoC project; it's not listed mostly
because we haven't had time to figure out what the natural next steps would
be (and because it's in a weird state until we can get the rest of your work
merged). You should some up with a careful project proposal so it's clear
how you're building on previous work. Some sort of dimensionality reduction
seems likely to be important, although there might be other approaches that pay
off as well. Having a strategy for (automated) performance analysis, and some
sort of reasonable targets, will be important here.

Given the state of the current PR, I think getting that merged should be
possible ahead of GSoC itself. What I suggest you do to make life easier is to
either use interactive rebase, or just build commits from scratch, on top of
latest master in order to address the concerns that were discussed in that PR.
The GSoC branch from last year only had one commit, and the PR is only really
two or three commits itself, so producing a clean PR against master should
result in something that can be reviewed and merged fairly quickly.

J

-- 
 James Aylett
 devfort.com — spacelog.org — tartarus.org/james/