Multi dimension minimization

Multi dimension minimization

[Emmanuel Chomarat]

  Hello,
  I  try  to  make  an app that find the max of the (log)likelihood of data that
  (  a  sum  of 75000terms = 5parameters*15000datas). I
  can't(don't)  want  to  use  the  derivates,  how  can  i do? Does anyone have
  something  to  do  it  in  a  box?  Does anyone heard about something in other
  numerical  lib(  numerical recipes have but i'm not sure to be able to use it,
  if someone have an sample).
  Thx for your help.

  Furthermore,  i  wrote  some mails about how to compile on win32 under borland
  env.  I think i succeeded ( i will write a little mail with the procedure) and
  i  have  to  test the libs, does anyone have a prog that make a lot of usage of
  GSL funcs.

-- 
 Emmanuel                          mail:Emmanuel.Chomarat@bigfoot.com

Re: Multi dimension minimization

[Brian Gough]

Emmanuel Chomarat writes:
 >
 >   Hello, I try to make an app that find the max of the
 >   (log)likelihood of data that ( a sum of 75000terms =
 >   5parameters*15000datas). I can't(don't) want to use the
 >   derivates, how can i do? Does anyone have something to do it in a
 >   box?  Does anyone heard about something in other numerical lib(
 >   numerical recipes have but i'm not sure to be able to use it, if
 >   someone have an sample).  Thx for your help.

The existing routines in GSL all use derivatives.  There is an
interface for minimization and fitting without derivatives but there
are no actual implementations yet.

 >   Furthermore, i wrote some mails about how to compile on win32
 >   under borland env.  I think i succeeded ( i will write a little
 >   mail with the procedure) and i have to test the libs, does anyone
 >   have a prog that make a lot of usage of GSL funcs.

The test programs can be compiled outside the build tree. 

RE: Multi dimension minimization

[J.J. Gomez Cadenas]

Hi,
Multidimensional minimization whithout derivates = Simplex algorithm. In
fact I use the simplex often even if I know the derivatives of the function.

It's a very simple algorithm. You can find the code in NR. I have my own
implementatin in C++ in case you want it.

As a matter of fact if it is not included in the GSL I would volunteer to
contribute with the code (I can write easily a C version)
jj

-----Mensaje original-----
De: gsl-discuss-owner@sources.redhat.com
[ mailto:gsl-discuss-owner@sources.redhat.com]En nombre de Emmanuel
Chomarat
Enviado el: martes, 21 de agosto de 2001 0:18
Para: gsl-discuss@sources.redhat.com
Asunto: Multi dimension minimization



  Hello,
  I  try  to  make  an app that find the max of the (log)likelihood of data
that
  (  a  sum  of 75000terms = 5parameters*15000datas). I
  can't(don't)  want  to  use  the  derivates,  how  can  i do? Does anyone
have
  something  to  do  it  in  a  box?  Does anyone heard about something in
other
  numerical  lib(  numerical recipes have but i'm not sure to be able to use
it,
  if someone have an sample).
  Thx for your help.

  Furthermore,  i  wrote  some mails about how to compile on win32 under
borland
  env.  I think i succeeded ( i will write a little mail with the procedure)
and
  i  have  to  test the libs, does anyone have a prog that make a lot of
usage of
  GSL funcs.

--
 Emmanuel                          mail:Emmanuel.Chomarat@bigfoot.com

RE: Multi dimension minimization

[Brian Gough]

J.J. Gomez Cadenas writes:
 > Hi,
 > Multidimensional minimization whithout derivates = Simplex algorithm. In
 > fact I use the simplex often even if I know the derivatives of the function.
 > 
 > It's a very simple algorithm. You can find the code in NR. I have my own
 > implementatin in C++ in case you want it.
 > 
 > As a matter of fact if it is not included in the GSL I would volunteer to
 > contribute with the code (I can write easily a C version)

Yes, that would be useful if you can plug it into the existing
frameworks for minimization/fitting.