[SciPy-User] Google Summer of Code...

[Sebastian Walter]

I could mentor an algorithmic differentiation (AD) project, but I'm
still not quite sure how to proceed.
Should I write up a proposal on the wiki and after that propose the
project on the
http://mail.python.org/mailman/listinfo/soc2010-mentors mailing list?

I have already created a package that I believe would be a very good
addition to scipy/numpy (preferably numpy) when it is mature enough.
It is hosted on http://github.com/b45ch1/taylorpoly

The package is not an AD tool but makes it very easy to create AD
tools using those algorithms.
An illustrative example how to do (forward mode) AD is given in:
http://github.com/b45ch1/taylorpoly/blob/master/examples/utps_for_algorithmic_differentiation.py

What is missing up to now are differentiated algorithms for tan, atan,
asin, sinh, etc. (in Taylor arithmetic)
as well as the core algorithms for reverse mode AD.

Implementing those algorithms should be doable in 4 months. Also,
since the task is to implement many small algorithms,
this project is unlikely to fail completely. The worst that can happen
is that not as many algorithms get implemented as one could wish for.

Could anyone give feedback if that project is ok?

best regards,
Sebastian





On Tue, Mar 2, 2010 at 12:32 AM, David Warde-Farley <dwf at cs.toronto.edu> wrote:
> On 1-Mar-10, at 5:59 PM, Skipper Seabold wrote:
>
>> I would also encourage anyone who has ideas to edit the wiki here to
>> reflect the current project wishlist (especially if you're willing to
>> mentor): http://projects.scipy.org/scipy/wiki/SummerofCodeIdeas  IMHO,
>> the more ideas there are, the more chance of grabbing interested and
>> diverse sets of students provided people are willing to mentor.
>
> I just reorganized the page a bit, to separate out the stuff that is
> already going on, like statsmodels, fwrap and the Py3k/datetime stuff,
> from things that have been listed for a while but never really picked
> up on.
>
> David
> _______________________________________________
> SciPy-User mailing list
> SciPy-User at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>