changes

Program in matlab for trainig rbf networks. I hope is good for you.

Here are the changes w.r.t. to the Scilab version of 
  Walter Di Carlo <[e-mail]> 03/04/99

User visible changes:

- The function has been renamed to  'DiffEvol'

- The function (to minimize) is no longer a string
  but the name of a Scilab function

- XVmin XVmax must be column vectors now

- The stepsize parameter F now has 2 components
  in case of strategies 3 and 8


Internal changes:

- The indices of all matrices have been reversed
  the first index is now 1:D
  This enables better vectorization

- All calls to Scilab's 'eval' (a relict of Matlab's feval)
  have been replaced by direct calls to the function (which is
  much faster)

- constructions like  ones(size(A)) where A is a matrix don't work
  in recent versions of Scilab; these have been replaced by
  ones(A)

- The separate function 'randperm' has been replaced by a call
  to Scilab's internal 'grand' function (which is faster)

- Methods 3 and 8 now use two different step sizes as with
  method DE2 in the paper http://www.icsi.berkeley.edu/~storn/deshort1.ps

- Some parameters (intput and output) have been renamed to better
  reflect their usage

- Tested with Scilab version 4.0

2006/03/01

Helmut Jarausch
RWTH-Aachen university
Germany

([e-mail])