[SciPy-user] constrained optimization

[eric jones]

> John Hunter wrote:
> 
> >>>>>>"Travis" == Travis Oliphant <oliphant.travis at ieee.org> writes:
> >>>>>>
> >>>>>>
> >
> >    Travis> It's been a while since I looked and I could very likely
> >    Travis> have missed some.  It's an area we would definitely like
> >    Travis> to add to scipy.
> >
> >I just spoke with one of the creators of optsolve, who I worked with
> >when I worked on that library, and told her about the interest at
> >scipy in a good optimizer library that could be extended to python,
> >and expressed my regrets about their 'free for non-commercial use'
> >license.  She said this license was currently under negotiation, and
> >I'm going to talk with them further about this possibility.  What
> >licensing terms does scipy require?
> >
> >If any you have any input for me about whether you think this would
be
> >a good inclusion for scipy if the license were right, give me some
> >feedback and I'll pass it their way.

Peter Stoltz is at Tech-X also.  I've copied him on the mail to give him
a heads up about the discussion.

> Libraries in SciPy need an unrestrictive open source license
(something
> like the Python license itself or the BSD license)  but Eric can
verify
> that.

That is right.

> 
> One thing we can also do is place the interface in SciPy but require
the
> user to download the library separately as an add on (we do that with
> the PIL right now for example)

PIL has a license that is compliant with SciPy.  fftw is another package
that we have optional support for that doesn't have a compatible
license.  The key here, though, is that there is a functional equivalent
in fftpack.  If fftw is used, SciPy doesn't loose any functionality, it
is just a little slower. 

For something as important as a constrained multi-dimensional optimizer,
the version we use needs to be a license compatible with SciPy (free for
academic/commercial, no restrictions that additions must be contributed
back to package).  I'm really interested in getting a good version of
one of these, but am willing to wait until we find one with a conforming
license, or someone develops one for SciPy.

See ya,
eric