[SciPy-Dev] testbench for some non-derivative optimizers in NLopt and scipy.optimize

Ralf Gommers ralf.gommers@googlemail....
Tue Apr 17 16:28:59 CDT 2012


On Thu, Apr 12, 2012 at 7:04 PM, denis <denis-bz-py@t-online.de> wrote:

> Folks,
>  fwiw, https://github.com/denis-bz/opt has testbenches
> for some of the non-derivative optimizers in NLopt and scipy.optimize.
> >From opt-py.Readme --
>
> These few test functions are *very* noisy, sensitive to random seed:
>    fmin    0.4      neval  369  LN_BOBYQA
>    fmin    3        neval  580  LN_BOBYQA
>    fmin    0.06     neval  423  LN_BOBYQA
>    fmin   12        neval  385  LN_BOBYQA
>    fmin    2        neval  325  LN_BOBYQA
> ...
>
> ==> we need test functions.
>
>
> Comments, any birds of a feather ?
>

This seems to be telling me something interesting, but not sure exactly
what:
https://github.com/denis-bz/opt/blob/master/nlopt/test/Powellsincos-dim10.png

It looks like there are a lot more tests for NLopt than for scipy.optimize.
Are you planning to keep developing this test bench?

Ralf
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.scipy.org/pipermail/scipy-dev/attachments/20120417/3c0a0307/attachment.html 


More information about the SciPy-Dev mailing list