[Xapian-devel] GSoC 2014

[Olly Betts]

On Mon, Feb 17, 2014 at 11:24:58PM +0530, Nikhar Agrawal wrote:
> On Sat, Feb 15, 2014 at 2:34 PM, Olly Betts <olly at survex.com> wrote:
> > Did you have an idea what you might want to work on?
> 
> Yes, the projects 'Weighing Schemes' and 'Learning to Rank' both seem
> interesting with my inclination being more towards 'Weighing Schemes'.
> What is higher priority for Xapian?

We'd much rather students chose the projects which they found most
interesting.

We don't really have a single priority order for the project ideas
anyway - different people will have different ideas of what matters.
All of the projects ideas have attached mentors, so they are all things
that at least one person would like to see worked on.

The only real consideration is that we don't want multiple students
working on the same thing (while GSoC rules do actually allow it, it
seems too problematic to us).  Note that it's totally possible for two
students to take the same project idea in different directions such that
there's little or no overlap, but if two students have submitted great
proposals with a lot of overlap, we'll discuss how to resolve this
rather than simply rejecting a great applicant.

But generally the students who start earlier end up submitting better
proposals, and students arriving later can see in the mailing list
archives which project ideas have attracted interest already and tend to
focus on others, so things tend to even themselves out.

> What would you like me to do next?

We'll be asking applicants to come up with a patch to help us see how
well they can dig into and work with the code, and to help us decide if
they've scoped their projects suitably.  So to get a head start on that
process, I'd suggest digging into the code and trying to come up with a
patch which solves a bug or implements a (probably small) new feature.  

There are some "bite-site" project ideas here:

http://trac.xapian.org/wiki/ProjectIdeas

You can also just browse the bugs in trac:

http://trac.xapian.org/report/1

Or work on a first step of your project, which can be a good way to get
a better feel for what's needed.  E.g. for weighting schemes, you could
look at adding support for another SMART normalisation code to
TfIdfWeight.  Even if you don't actually get it fully working, trying
to do it will give you a better understanding of the existing code
relating to weighting schemes.

If you have questions or need a hand getting going, we're here to assist
(either here or IRC - mail's better for complex questions, IRC's better
for interaction, though timezones can be an issue).

Cheers,
    Olly