leasqr: error: weighted residuals are not real

[oxy]

Hi Guys,

this is an old problem. I always spot it again.

There are old email threads about it. They say it is a bug.

I could not find discussions since 2012.

Is it realy a bug or am i doing sth wrong?

working with Octave, version 4.0.1, optim Version: 1.5.0
Thanks very much in advance!


%====== here a code example ================


clear all

M=[
   5.3900e-01   0.0000e+00
   1.0600e+00   1.8838e-01
   1.5730e+00   2.7787e-01
   2.0900e+00   3.3691e-01
   2.6070e+00   3.8265e-01
   3.1240e+00   4.1204e-01
   3.6410e+00   4.3912e-01
   4.1580e+00   4.6368e-01
   4.6750e+00   4.8223e-01
   5.1920e+00   4.9765e-01
   5.7080e+00   5.1288e-01
   6.2250e+00   5.2316e-01
   6.7420e+00   5.3964e-01
   7.2590e+00   5.4866e-01
   7.8070e+00   5.6333e-01
   8.3240e+00   5.7266e-01
   8.8410e+00   5.8163e-01
   9.3580e+00   5.9472e-01
   9.8750e+00   6.0218e-01
   1.0392e+01   6.0394e-01
   1.0959e+01   6.1534e-01
   1.1473e+01   6.2155e-01
   1.1990e+01   6.2744e-01
   1.2507e+01   6.3283e-01
   1.3023e+01   6.3878e-01
   1.3540e+01   6.4856e-01
   1.4061e+01   6.5112e-01
   1.4574e+01   6.5670e-01
   1.5091e+01   6.6284e-01
   1.5608e+01   6.6585e-01
   1.6125e+01   6.7262e-01
   1.6642e+01   6.7582e-01
   1.7159e+01   6.8428e-01
   1.7676e+01   6.8691e-01
   1.8193e+01   6.9230e-01
   ];

   Mt=@(t,p) p(1) * (p(2) - p(3)*exp(-((p(4)*(t+p(5))).^p(6))));
   pin=[1.1;   0.9;   0.7;   0.1;   0;   0.6];
   stol=0.0000001;
   niter=10000;
   wts = (1/sqrt(M(:,2)))';
   dFdp='dfdp';
   dp=0.001 * ones (size (pin));

   options.bounds=[0.5    1.5; 0.5    1.5 ;  0.5    1.5 ; 1e-16
1e16; -10   50;   0.1    1.2  ];
   [f, pfound, cvg, iter]=leasqr(M(:,1), M(:,2), pin, Mt, stol, niter
,wts, dp, dFdp, options);