Doesn't sound like you need a mentor :-) I'd just start by picking
something you are interested in and is useful for you and work on it
and submit a patch. Consider the community to be the mentor. Just
feel free to ask questions and put up patches. Patches don't have to
be perfect, they just need to be followed up on.
I've done some record linkage, so I will be interested to check out
your approach w/ EM.
-Grant
On Mar 30, 2008, at 1:21 AM, sarp wrote:
Summary:
I'm very excited about Mahout project and my goal is to become one
of the
committers for this project. To start with, I'm willing to implement
one of
the algorithms that won't be covered by GSoC students. What I'm
looking for
is a mentor who would be willing to supervise me like a GSoC
student, even
though I won't be one, and help me in formulating these algorithms
in the
Map/Reduce framework, as well as provide feedback about
design/implementation issues I will face during this process.
About Me:
I'm a Fulbright scholar from Turkey, studying at Georgia Institute of
Technology for a M.S. degree in CS with a focus on machine learning.
I've
successfully participated in GSoC last year [1], and worked with
OpenMRS
organization on "Patient Matching and Record Linkage" project [2].
We used
EM algorithm for estimating parameters in our statistical model, in
order to
predict whether two records belong to the same patient or not. I
will be
doing an internship during summer, so I thought it would be fair to
give
others who have more time on their hands a chance to experience
GSoC, that's
why I won't be applying this time.
My Background:
I've taken courses on Statistical Modeling, Data Mining [3], Machine
Learning [4] and Computational Data Analysis [5], so I have the
necessary
background to implement most of the algorithms mentioned in the
original
paper [6]. Most of my experience is regarding unsupervised learning,
I've
implemented k-means and hiearchical clustering before, and currently
I am
working on implementing k-medoids and constrained k-means algorithms
[7] in
C++ for FASTlib (A library of Fundamental Algorithmic and
Statistical Tools)
[8]. My motivation for participating in this project is to enhance my
understanding of ML algorithms by implementing them, learn about Map/
Reduce
framework, and possibly make use of Mahout in my MS research project.
Links:
[1]
http://googlesummerofcode.blogspot.com/2007/12/friday-fulbright-and-rhodes.html
[2]
http://code.google.com/soc/2007/openmrs/appinfo.html?csaid=E680200FD32E82D6
[3] http://www2.isye.gatech.edu/~shan/ISyE7406/ISYE7406.html
[4]
http://www-static.cc.gatech.edu/fac/Charles.Isbell/classes/2008/cs7641_spring/syllabus.html
[5] http://www.cc.gatech.edu/~agray/spr08.html
[6]
http://www.cs.stanford.edu/people/ang/papers/nips06-mapreducemulticore.pdf
[7] http://www.litech.org/~wkiri/Papers/wagstaff-kmeans-01.pdf
[8] http://www.cc.gatech.edu/~agray/fastlib.pdf
Grant Ingersoll-6 wrote:
Just thought I would mention a couple of things about GSOC.
First, thanks to all for the interest. It is great to see ML alive
and well in universities. We have several good applications already.
Since the ASF is only alloted a certain number of picks, we go
through
an internal ranking process to decide which projects get them (they
try to spread them out over many projects). So, while we have 4
willing mentors, it doesn't necessarily mean we will get that many
students.
However, to go along with that I would encourage all students to go
through their proposals and make sure they can fill in details about
their plans as much as possible, as well as their bios, etc. Also,
please don't be shy about discussing your plans here. One of the
requirements of doing this project is going to be to interact with
your mentor and the community.
Finally, I would certainly like to encourage those who don't get
selected to stick around and contribute. I am sure others in the
community can vouch for this: being an active contributor/committer
to
a project like this is a real edge when it comes to getting a job
doing what you want to do. Think about being in a job interview for a
ML company and saying, yeah: I contributed algorithm X to Mahout
which
was used to by Y on a 100 node cluster, or I contributed these 10
patches to Mahout plus I'm an active discussion participant, go look
it up. It gives potential employers an incredible track record to
review and shows you know how to get along with others in a "work"
environment.
At any rate, enough of the pep talk. Good luck to you all, I look
forward to evaluating the ideas!
Cheers,
Grant
--
View this message in context:
http://www.nabble.com/Thoughts-on-GSOC-tp16369329p16378749.html
Sent from the Mahout Developer List mailing list archive at
Nabble.com.
--------------------------
Grant Ingersoll
http://www.lucenebootcamp.com
Next Training: April 7, 2008 at ApacheCon Europe in Amsterdam
Lucene Helpful Hints:
http://wiki.apache.org/lucene-java/BasicsOfPerformance
http://wiki.apache.org/lucene-java/LuceneFAQ
