[OctDev] Re: small change to leasqr.m

[Etienne G. <et...@is...>]

  Hello,

alternatively, you could do as in the other optimization functions and
allow 'pin' (or 'x') in

    leasqr(x,y,pin,F,{stol,niter,wt,dp,dFdp,options})

to be a list : pin = list (p1,p2,...). 'F' would be called with x, p1,
p2 ... as arguments and p1 is the adjusted argument. I don't know how
this fits w/ matlab's syntax. I guess a patch that doesn't break
anything would be accepted.

  Cheers,

  Etienne

On Wed, Aug 21, 2002 at 09:12:13AM -0600, E. Joshua Rigler wrote:
# I want to recommend, and perhaps eventually submit a small change to the
# leasqr.m function in octave-forge that should be completely transparent
# and backward compatible.  Feedback on my idea, and instructions for
# submitting this change would be appreciated.
# 
# In the eval calls to the function "F" (i.e. the function who's
# parameters are being estimated), and "dFdp" (either the numerical, or
# user-written analytic derivative function), I suggest that a structure
# is passed that contains miscellaneous information that is neither an
# input or output vector, nor one of the parameters.  It will be the job
# of the user-written function to use the information in that structure
# appropriately.
# 
# This, of course, will require an additional input to leasqr.m, but if
# the structure is placed as the last input variable, no previously
# written code should be affected.  It also seems to me that this should
# be compatible with Matlab, which seems to have been another goal of the
# authors of leasqr.m.
# 
# My only option now (aside from changing my local copy of leasqr.m) is to
# set a global variable, or variables, that define the options I want, but
# I always feel a little tainted using global variables (a consequence of
# working on too much bad FORTRAN code I suppose).  If someone has a
# better option, please let me know.  Thanks.
# 
# -EJR


-- 
Etienne Grossmann ------ http://www.isr.ist.utl.pt/~etienne