[SciPy-dev] [scikits] openopt SVN instable for the moment

Matthieu Brucher matthieu.brucher@gmail....
Wed Apr 9 01:01:55 CDT 2008


>
> On Tue, 08 Apr 2008, dmitrey wrote:
> > I still don't think it's a good idea
>
>
> Are you two going to work this out in private or are you
> going to add some details about the design considerations
> and seek input from the SciPy community?
>

I'd like to have some feedback. For the moment, the changes will not be seen
by the typical end-user. No API was changed.
For the sys.path modification to be removed, I had to find a way of
importing the corresponding solver that is stored somewhere in the
hierarchy, generally in a module named scikits.openopt.solvers.*.solver_oo.
So now that the parent modules (scikits.openopt.solvers.*) are no longer in
the path, the easiest way (at least for me) was to add in a registry this
information, so the available_solvers registry was created.
In this configuration, everything that is in solvers uses the Kernel, but
the Kernel does not use anything in solvers, they are only called by the
registry. I think that dmitrey did a great job in having a clean cut here.

For the moment, the registry is not advertised and accessible by
scikits.openopt.available_solvers. It is still hidden in
scikits.openopt.Kernel.runProbSolver, but this can be easilly modified.
Now people can register a new wrapper by calling :

scikits.openopt.Kernel.runProbSolver.available_solvers['name'] =
'path.to.module'

I agree with dmitrey, there will not be a lot of people that will use this,
but if we can get one contributed wrapper this way, I think it will be great
for everyone, dmitrey included, as he is the one behind the current
wrappers.

On a side note, the generic framework is finally importable like any other
scikit (this was not the case until 2 or three days ago, becuse of this
design, and the Technology Preview in scikits.learn can heavily depend on
it) :
from openopt.scikits.solvers import optimizers

So the available_solver is the only addition in the scikit, and I think that
wrappers developers will appreciate not having to add files to the scikit.

Matthieu
-- 
French PhD student
Website : http://matthieu-brucher.developpez.com/
Blogs : http://matt.eifelle.com and http://blog.developpez.com/?blog=92
LinkedIn : http://www.linkedin.com/in/matthieubrucher
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.scipy.org/pipermail/scipy-dev/attachments/20080409/606b1fab/attachment.html 


More information about the Scipy-dev mailing list