[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


From: etienne grossmann
Subject: downsimplex
Date: Tue, 18 Jan 2000 16:29:46 +0000 (WET)

  Hello Rolf,

#  Hi, 
# we all know and regret, that octave's optimization capabilaties are
# rather limited, so all contributions like your downsimplex
# minimization or Levenberg Marquard approaches are really welcome.
#  Especially in connection to a lacking ,fmins' analogon, which is
#  quite often a point of discussion in the ,help-octave' list, you're
#  recommending your downsimplex program sometimes..
#  After playing around with ,downsimplex' a while (about 1/2 hour),
#  let me give some comments: According to my opinion, the way of it's
#  argument handling is much too complicated and prevents it (most
#  propably) from being widely used.

  Unfortunately, I agree entirely. What kind of synopsis would you

#  OK, there's a description of the arguments in the header, but in
#  order to understand them including their types and allowed
#  combinations of optinonal and/or mandatory arguments, it's
#  necessary to have a very close look at the code itself.
#  I strongly suggest to add some 
#  really SIMPLE examples as a GUIDELINE of usage
#  to the header, illustrating the use of the most important
#  options.


#  E.g. the test I played around with was simply
#  finding the unique minimum of the scalar test-function
#  y = x^x         on the positive real axis. 
#  which occurs at xmin=1/e and corresponding ymin=(1/e)^(1/e) ~=0.69220...
#  Can you please tell me how to use ,downsimplex' in this simple case
#   to get these results ? 

  Sure :

octave:13> function y = dummyfunc(x), y = x^x ; end
octave:14> [x0,val,niter]=downsimplex("dummyfunc",0)
x0 = 0.36788
val = 0.69220
niter = 38

  BTW, I just noticed (it's a bug) that I couldn't minimize a function
called "f" because of name collision ... 

  If you like, we should be able to write decent minimizing tools. It
should be easier for 2 persons than for one.

#  Thanks  Rolf


Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:
How to fund new projects:
Subscription information:

reply via email to

[Prev in Thread] Current Thread [Next in Thread]