[scilab-Users] Problems of compilation of nightly build under opensuse
Michaël Baudin
michael.baudin at scilab.org
Mon Oct 19 10:49:21 CEST 2009
Dear Sébastien,
Thank you for reporting your experience with fminsearch
in the nighty build.
Any bug which might be discovered before the actual release
of v5.2 will help us to release an even more robust version.
As soon as you are able to run any version of Scilab (be it a
nighty build or a source version), whatever the platform,
do not hesitate to report any bug you may find on our
bug tracking system :
http://bugzilla.scilab.org
Could you clarify the error message you got with respect
to the "funccount vs funcCount" problem ?
Your problem with the number of function evaluations is not clear for
me. I checked the behaviour of Scilab's fminsearch by comparison
with Matlab's, and found exactly the same number of function
evaluations in the end (I can provide further details if necessary).
The initial number of function evaluations is n+1,
since one function evaluation is required for each vertex.
Best regards,
Michaël
Sébastien Bihorel a écrit :
> By the way,
>
> If that can help anybody finalizing the Nelder-Mead algorithm, I saw
> something that looks like a bug to me. Although I was unable to run
> the fminsearch function to its end, I saw that for the initial
> iteration (ie. the iteration 0), more than one function evaluations
> are reported... this is weird, the cost or objective function (however
> your want to call it) should only be evaluated once, ie using the
> initial guesses of the parameters to optimize.
>
> Sebastien
More information about the users
mailing list