Re: leasqr question (again)

[Olaf Till]

On Fri, Sep 14, 2012 at 11:10:48PM +0200, Olaf Till wrote:
> On Fri, Sep 14, 2012 at 12:44:08PM -0700, seektime wrote:
> > Hi,
> > 
> > I'd like to run leasqr.m in its simplest form, ie without any options (just
> > bare bones) but I'm getting the following error message (variables pasted
> > below if you want to test it on your system):
> > 
> > octave:3> [f,p]=leasqr(V,I,pin,F)
> > error: Invalid call to options.  Correct usage is:
> > 
> >  -- Function File: OPT = options ("KEY1", VALUE1, "KEY2", VALUE2, ...)
> > 
> > error: called from:
> > error:  
> > C:\octave\Octave3.6.1_gcc4.6.2\share\octave\3.6.1\m\help\print_usage.m at
> > line 87, column 5
> > error:  
> > C:\octave\Octave3.6.1_gcc4.6.2\share\octave\packages\control-2.3.52\options.m
> > at line 68, column 5
> > error: evaluating argument list element number 1
> > error:  
> > C:\octave\Octave3.6.1_gcc4.6.2\share\octave\packages\optim-1.2.0\leasqr.m at
> > line 574, column 5
> > 
> > Code and variables are:
> > 
> > 
> > V=[
> >  -159.88
> >  -149.27
> >   -139.1
> >  -129.08
> >  -118.97
> >   -113.9
> >  -108.86
> >  -103.82
> >  -98.762
> >  -93.748
> >  -88.717
> >  -83.669
> >  -78.658
> >  -68.557
> >  -58.494
> >  -48.459];
> >  
> >  I=[
> >        1
> >   1.0079
> >  0.97652
> >  0.87318
> >  0.81173
> >  0.72704
> >  0.66648
> >  0.57694
> >  0.39872
> >  0.29661
> >  0.18571
> > 0.094664
> > 0.047699
> > 0.035285
> > 0.021977
> > 0.017704];
> > 
> > pin=([-100 -10]);
> > F = @(V,p) 1 ./ (1 + exp((p(1) - V) / p(2)));
> > 
> > Besides that fact that it doesn't return the parameters I'm wondering about
> > the message since I'm not passing any options? Any hints, explanations or
> > suggestions would be highly appreciated.
> > 
> > Michael
> 
> Oh, this is a bug ... I put in some testing code some time ago and was
> not careful enough.
> 
> The line 574 in leasqr.m
> 
>     if (isfield (options, 'testing'))
> 
> should acutally be
> 
>     if (nargin > 9)
> 
> As it is now, if no argument 'options' is given, the test is also
> positive if some function 'options' is provided anywhere --- in your
> case within the control package. This function is then erroneously
> called and gives the error message you have seen.
> 
> Sorry. I'll correct this in SVN, and I think a bugfix release of
> 'optim' is justified in this case. In the meantime, you could make the
> above change in line 574 yourself, or just delete the whole block
> commented as %% only preliminary, for testing.
> 
> Olaf

I should also say that if you can't wait for the fixed package you
could use a different function. With the variables given by you and
global "verbose" not set,

[f, p] = leasqr (V, I, pin, F)

is equivalent to (same core algorithm):

[p, f] = nonlin_curvefit (F2, pin.', V, I)

if F2 is defined as F but with different order of arguments:

F2 = @ (p, V) 1 ./ (1 + exp ((p(1) - V) / p(2)));

Note that nonlin_curvefit requires the initial values to be a column
vector (if they are given as a vector).

Olaf

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: Digital signature