Re: help needed with nonlinear inequality constraints (optimset "inequc") for nonlin_residmin in optim package

[Olaf Till]

On Fri, Feb 16, 2018 at 12:49:23PM +0100, Philippe Dobbelaere wrote:
> settings = optimset (settings, "inequc", address@hidden);
> 
> with
> 
> function H = nonlinear_ineq_constraints(x)
>   h0 = 1 - get_ic0(x)^2 - get_jc0(x)^2 - get_kc0(x)^2; # >= 0
>   h1 = 1 - get_ic1(x)^2 - get_jc1(x)^2 - get_kc1(x)^2; # >= 0
>   h2 = 1 - get_ic2(x)^2 - get_jc2(x)^2 - get_kc2(x)^2; # >= 0
>   H = [h0;h1;h2];
> endfunction
> 
> error: number of columns must match
> error: called from
>     __nonlin_residmin__> at line -1 column -1
>     __nonlin_residmin__> at line -1 column -1
>     __nonlin_residmin__> at line -1 column -1
>     __lm_svd__ at line 181 column 6
>     __nonlin_residmin__ at line 1133 column 21
>     nonlin_residmin at line 98 column 21

The probable cause was a bug, simply a missing line continuation '...'
in line 970 of inst/private/__nonlin_residmin__.m . The fix is
pushed. Can you use it without a new release?

This shows that I really must include tests to prevent such things
from creeping in... I had not even a demo included for nonlinear
constraints...

BTW: In this case the problem was easy to reproduce, but in general
you should provide a minimal code for demonstration. And a newer
Octave version would also help, giving line numbers in anonymous
function definitions, not just:

>     __nonlin_residmin__> at line -1 column -1
>     __nonlin_residmin__> at line -1 column -1
> ...

Olaf

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

signature.asc
Description: PGP signature