nonlinear optimization functions

Librairie C++ de calcul formel/ C++ symbolic computation library

Modérateur : xcasadmin

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » jeu. janv. 24, 2019 5:03 pm

The culprit seems to be the identifier a, since the command below works as expected for me.

Code : Tout sélectionner

for k from 1 to 100 do extrema(x*y*z,x^2+y^2+z^2=1,[x,y,z]); od:;
I'ts probably something related to my system. We'll see if someone else experiences the crash.

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » sam. févr. 02, 2019 6:34 pm

I had to restore euler_lagrange and jacobi_equation commands, I've accidentaly deleted the implementations some time ago... Now optimization.cc is complete again. Perhaps it happened during slicing the file when moving kovacicsols. Fix committed to my source.

parisse
Messages : 5734
Inscription : mar. déc. 20, 2005 4:02 pm
Contact :

Re: nonlinear optimization functions

Message par parisse » dim. févr. 03, 2019 1:37 pm

Indeed!

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » dim. févr. 03, 2019 11:50 pm

I had to restore convex also. I updated optimization.cc again and verified all commands from optimization.h one by one to be sure I haven't missed something. Seems OK now. Sorry about that!

parisse
Messages : 5734
Inscription : mar. déc. 20, 2005 4:02 pm
Contact :

Re: nonlinear optimization functions

Message par parisse » mer. févr. 06, 2019 2:05 pm

Ok, source synchronized on my side. I hope I can do a new release soon, I'm still trying to fix something related to a gbasis optimization.

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » jeu. févr. 07, 2019 9:02 pm

There are also a couple of corrections in the graphtheory manual I committed today.
By the way, I installed Xubuntu 18 and compiled giac using your source tarball from today with fltk 1.3 and build-essential without problems. Everything works fine, including nauty and GLPK, and there are no problems with extrema(a*x*y*z,x^2+y^2+z^2=1,[x,y,z]) commandline. However, the problem persists here with version 1.5.0-37 installed from the deb file of Jan 14.

parisse
Messages : 5734
Inscription : mar. déc. 20, 2005 4:02 pm
Contact :

Re: nonlinear optimization functions

Message par parisse » ven. févr. 08, 2019 12:08 pm

doc updated on my side. I'll make a new debian on xubuntu 18, and I'll check.
By the way, there was a massive change in all your file permissions : mode change 100644 => 100755

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » ven. févr. 08, 2019 12:50 pm

Indeed, the permission changes happened because the files were moved to another partition via a backup drive (I should've had simply make a pull). I changed the permissions for files back to 0644 and set git to ignore such changes in the future.
The latest commit includes a new command in signal processing, moving_average filter. I'll update the docs today. Should I to synchronize cascmd_en.tex and/or aide_cas before that?

parisse
Messages : 5734
Inscription : mar. déc. 20, 2005 4:02 pm
Contact :

Re: nonlinear optimization functions

Message par parisse » ven. févr. 08, 2019 2:54 pm

Yes for aide_cas, I have added total_degree.

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » ven. févr. 08, 2019 7:49 pm

Documentation updated! There are two new images in the doc directory.
[Edit]
I also moved implicitdiff to "Derivation and applications" section in cascmd_en.tex.

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » ven. févr. 22, 2019 8:55 am

Can't reproduce the crash with extrema(x*y*z,x^2+y^2+z^2=1,[x,y,z]) on ubuntu in 1.5.0-43. It's working fine now.
By the way, there is incompatibility between giac and ubuntu 18.04, I had to install couple of deb packages by hand (libreadline6, for example). That was on a clean ubuntu install with giac stable repository enabled.

parisse
Messages : 5734
Inscription : mar. déc. 20, 2005 4:02 pm
Contact :

Re: nonlinear optimization functions

Message par parisse » ven. févr. 22, 2019 2:18 pm

There is a specific deb package for Ubuntu 18, cf. https://www-fourier.ujf-grenoble.fr/~pa ... n#packages

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » mar. févr. 26, 2019 5:15 pm

Thanks! Do you perhaps have deb files compiled with nauty and GLPK?
Btw, I fixed couple of bugs in optimization.cc today. The command extrema has been having some problems with critical point classification which produced incorrect or not well defined results in some cases. Also, minimax was crashing on some examples, which is now fixed. I also made small corrections in cascmd_en.tex and aide_cas regarding the examples. Graphtheory manual is also updated.

[Edit] I fixed another small bug in extrema command, now epsilon2zero is applied to the list of critical points before classification.

lukamar
Messages : 331
Inscription : ven. juin 30, 2017 9:55 am
Localisation : Zagreb, Croatia

Re: nonlinear optimization functions

Message par lukamar » dim. juil. 14, 2019 1:46 pm

Hello Bernard,

I have improved extrema finding algorithms. Some bugs are fixed: in particular, critical points are now discarded when undefined or non-constant, which suppresses errors and stalling on some examples, and finding images of homogeneous polynomials is improved. I committed the fixes today in my repository. There are also few corrections in the graph theory manual.

[Edit] I removed second-order derivative test by computing definitness of the Hessian via eigenvalues, it is actually not necessary and gives me errors for some examples.

parisse
Messages : 5734
Inscription : mar. déc. 20, 2005 4:02 pm
Contact :

Re: nonlinear optimization functions

Message par parisse » mer. juil. 17, 2019 4:11 pm

Thank you! I have updated my source codebase, but I won't be at my University during the next weeks, the next update should only happen during the first week of September.

Répondre