[Octave-bug-tracker] [bug #65535] (optim) lsqnonlin, lsqcurvefit: non-scalar resnorm, missing lagrange multipliers for both l

[A.R. Burgers]

URL:
  <https://savannah.gnu.org/bugs/?65535>

                 Summary: (optim) lsqnonlin, lsqcurvefit: non-scalar resnorm,
missing lagrange multipliers for both l
                   Group: GNU Octave
               Submitter: arb
               Submitted: Fri 29 Mar 2024 03:30:02 PM CET
                Category: Octave Package
                Severity: 3 - Normal
                Priority: 5 - Normal
              Item Group: Incorrect Result
                  Status: None
             Assigned to: None
         Originator Name: 
        Originator Email: 
             Open/Closed: Open
                 Release: dev
         Discussion Lock: Any
        Operating System: Any
           Fixed Release: None
         Planned Release: None


    _______________________________________________________

Follow-up Comments:


-------------------------------------------------------
Date: Fri 29 Mar 2024 03:30:02 PM CET By: A.R. Burgers <arb>
There are inconsistencies between matlab and optim in the lsqnonlin and
lsqcurvefit functions.

The 1st issue is that for multi-column ydata a non-scalar resnorm is
returned.

The 2nd issue is that the Lagrange multiplier output lambda is empty.

octave 10, optim-1.6.2 output:


lsqnonlin
resnorm =   0.020000   0.048223
lambda = [](0x0)

lsqcurvefit
resnorm =    0.020000   0.048223
lambda = [](0x0)


matlab R2019b output


resnorm = 0.1759
lambda = struct with fields:
    lower: 0
    upper: 5.2781e-05

lsqcurvefit
resnorm = 0.1759
lambda = struct with fields:
    lower: 0
    upper: 5.2781e-05


test script:


if exist('OCTAVE_VERSION', 'builtin') > 0
  pkg load optim;
end
rng('default'); % for reproducibility
xdata = reshape(linspace(0,3,40), [], 2);
ydata = exp(-1.3*xdata) + 0.05*randn(size(xdata));
x0 = 2; lb = 1; ub = 3;
opts.Display='off';

fprintf('lsqnonlin\n');
fun_lsqnonlin = @(r)exp(-xdata*r) -ydata;
[x,resnorm,residual,exitflag,output,lambda] = ...
     lsqnonlin(fun_lsqnonlin,x0, lb, ub, opts);
resnorm
lambda

fprintf('lsqcurvefit\n');
fun_lsqcurvefit = @(r, x) exp(-x*r);
[x,resnorm,residual,exitflag,output,lambda] = ...
     lsqcurvefit(fun_lsqcurvefit, x0, xdata, ydata, lb, ub, opts);
resnorm
lambda


The lambda field is correctly calculated, but unintentionally cleared because
I think the user_interaction detection in this snippet around line 283 is
incorrect, since there is no use_interaction in this example to my
understanding.


  if (out_args >= 5)
    outp = residmin_out{4};
    if (isfield (outp, "user_interaction")) # not set if stopped by
user_interaction
      outp = rmfield (outp, "lambda");
    endif
    if (isfield (outp, "user_interaction"))
      outp = rmfield (outp, "user_interaction");
    endif
    varargout{5} = outp;
  endif









    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?65535>

_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/