[SciPy-user] in search of a different fmin method

Robert Kern rkern at ucsd.edu
Mon Aug 1 05:50:24 CDT 2005


Victor Reijs wrote:
> Hello all of you,
> 
> I now have scipy working for PSP8 (based on python 2.2) and I use fmin 
> (or fmin_Powell); the results are as if these method only find a local 
> minimum and not the global minimum.
> 
> I have the same problem programmed in MS Excel 2000 and using Excel 
> Solver it looks from the result that the global minimum is found (MS 
> Excel 2000 uses Generalized Reduced Gradient Algorithm: 
> http://support.microsoft.com/kb/q214115/ ).

This algorithm is not a global optimizer. See the section "Multiple 
Locally Optimum Points" on that page.

> Do people know if such a fmin is also available in Python?

We don't have many global optimizers. There's a simulated annealing 
implementation in scipy.optimize.anneal(). There's also a genetic 
algorithm implementation in scipy.ga, but I think it may have succumbed 
to bit-rot by now.

Global optimization is hard in general, though, and it takes significant 
experience with the algorithm to tune it for your problem. Good luck!

-- 
Robert Kern
rkern at ucsd.edu

"In the fields of hell where the grass grows high
  Are the graves of dreams allowed to die."
   -- Richard Harter



More information about the SciPy-user mailing list