[SciPy-user] Would anyone connect fortran constrained linear least squares solver to Python?

dmitrey dmitrey.kroshko@scipy....
Mon Jan 7 11:17:31 CST 2008


Hi Dominique,

openopt development is a little bit frozen for now, since I still search 
for a money support - I haven't got any since GSoC 2007 finish, but I 
try to remain active as long as possible (as my mentor Alan G Isaac 
recommended me), waiting for another one GSoC  (according to rules I 
should remain being student at ~April 11 and I intend to do my 
graduation after the date, to take participation in GSoC 2008).

So all I do is eating low-hanging fruits: either connecting 
well-documented solvers with python bridge already provided by someone, 
or doing some other changes (minor ones, I have no possibilities to 
start big jobs).

Unfortunately, NLPy still lacks precise convenient documentation. Would 
you provide something like the one:
http://projects.scipy.org/scipy/scikits/browser/trunk/openopt/scikits/openopt/examples/llsp_1.py
then I could provide openopt binding for the solver.

As for the netlib routine mentioned, of course I would gladly connect it 
would someone provide fortran-python bridge for the routine. Taking into 
account I have no experience of using f2py, SWIG, ctypes and other 
similar Python <-> xxx bridges, I have neither possibility no willing to 
spend my time and efforts for to provide it by myself.
Moreover, OO is 100% python-written (to prevent possible cross-platform 
with installation) and fortran routines should go to scipy and/or other 
software already containing C/Fortran code.

BTW, an OpenOpt user from Princeton informed me of willing to contribute 
something, and then we stopped at connecting IPOPT (to OO), so 2-3 month 
later (I hope) OO users will have IPOPT binding as well.

Best regards, Dmitrey

Dominique Orban wrote:
> On Dec 26, 2007 12:29 PM, dmitrey <dmitrey.kroshko@scipy.org> wrote:
>   
>> Hi all,
>>
>> I had noticed (from traffic statistics): lots of people are interested
>> in linear least squares problems (LLSP). However, scipy has only
>> unconstrained LAPACK dGELSS/sGELSS.
>>
>> Could anyone provide connection of the fortran-written solver to Python
>> (or connect it to scipy)?
>>
>> http://netlib3.cs.utk.edu/toms/587
>> (that one can handle linear eq and ineq constraints)
>>
>> Then I would gladly provide connection of the one to scikits.openopt.
>>     
>
> Hi Dmitrey,
>
> Please note that NLPy now features a pure Python version of Michael
> Saunders' LSQR for unconstrained linear least-squares:
> http://nlpy.sf.net
>
> Regarding the Netlib Fortran code, can't we use f2py?
>
> Cheers,
> Dominique
>   



More information about the SciPy-user mailing list