Re: Matlab central and octave

[David Bateman]

address@hidden wrote:
> So I took a look at the link http://octave.dbateman.org (2 packages - optim &
> parallel).
> First of all, am I right that all your Octave optimization toolbox consist
> just from 1 function - gls?
> This seems to be so when I see your link
> http://www.gnu.org/software/octave/doc/interpreter/Optimization.html#Optimization
> 3 subsections -
> # Linear Programming
> # Quadratic Programming
> # Nonlinear Programming 

Try in

http://octave.dbateman.org/doc/optimization.html

and in

http://octave.dbateman.org/doc/optiminterp.html

is probably more what you want to do as that will include the octave
core optimization functions as well

> are empty and only 1 contains of gls. (I was seeking MATLAB linprog() analog
> for my needs). So, am I right or wrong?
> Next, at http://octave.dbateman.org I counted 5-6 solvers for nonsmooth
> unconstrained (UC) optimization and 7-8 for smooth unconstrained - and
> almost nothing for constrained (first of all I speak about fmincon analog).
> I would say this amount of UC solvers is funny if it wasn't so sad. Even
> MATLAB has just 1 UC smooth solver (fminunc), 1 constrained (fmincon), 1
> nonsmooth (fminsearch). 

Dmitry I personally couldn't be bother as the optimization I need done
can be done with the functions available. Improving the code is left to
the developer who has that itch to stratch. As this seems to be you I'd
go fix the code is a better use of your time than complaining about the
existing functions.


> Moreover, calling sintax differs too much, it requires for users to rewrite
> it very much, while trying these solvers. So, if users (as a rule, students
> with their homework) will send each month 1 more UC solver, will you include
> them to octave-forge? Even if it will be 0.5% better than all other? 
> What about parallel toolbox - 2 funcs from 9 are undocumented (+some
> undocumented funcs in optim). dfeval() analog - the firstest thing to
> implement - is absent (I mean with same calling syntax y= dfeval(fun, x)).

Fix it..

> So I'm absolutely agree with Soren hierarchy proposition, and that "it
> requires contact with other human beings, which can be scary and takes
> time." So users must initially prove octave-forge workers that their code is
> actual & well-done. But what, for example, about some aerospace of medicine
> m-file routines? who will check that they are essential & have enough
> quality to be not just students homework? Or contain some bugs, yielding
> wrong results? And what users should do - waste time for connecting their
> code to octave-forge routines, 

And this as opposed to correcting bugs in Mathworks code.. We all know
their code isn't perfect either.. This is the problem with relying on
some elses code and is not specific to octave or octave-forge. There can
be no guarentees and so knowing what you are doing and what is a sane
result is what matters.


> trying to determing, does it good enough to
> rely or no? And if there are some good ratings and reviews at matlab
> filexchange area, users download only those (from large amount of students
> homework) and are sure that it should be tried.
> Also, it must include system of ratings, reviews and feedback. It will
> demand some efforts for creating this web-interface only 1st time. Switching
> Octave programmists to the job is not nesesary - you may hire for very small
> money someone at GetAFreelancer.com or Scriptlance.com or any other
> freelancer's site - doesn't matter, they are connected with each other.
> WBR, D.

This thread can go nowhere if the person doing the work can't be
found... If Dmitry you said you wanted to setup "Octave central" and
support it I'd take a very different tone..

Regards
David