Re: Package for level-set-based shape optimisation

[Olaf Till]

Some answers from the perspective of the optim package:

On Wed, Mar 11, 2015 at 12:16:36PM +0100, Daniel Kraft wrote:
> Hi all!
> 
> I'm working on shape optimisation in a level-set framework.  Some parts
> of my code are already published in the Octave "level-set" package.  I
> would now like to also polish and publish some more, namely code related
> to the actual optimisation.
> 
> In particular, I'm thinking about the following parts:
> 
> 1) High-level routines that handle line searches and run a descent
> method on shapes described by level sets.

Could go into optim package if it's a general, established line search
algorithm which calls a user function (passed as argument) on its way,
assuming that all the extra stuff with the shapes can be done by this
user function (which I don't know). Otherwise, I'd agree it probably
wouldn't fit into optim package.

> 2) Related utilities for (optionally) plotting status information along
> the descent and also recording / replaying descents.  (Useful for
> running the computation on a server or over night and then interpreting
> the results interactively.)

See above, can this be done by a user function passed to the line
search? In any case, this code should probably not go into the optim
package (as you already said).

> 3) Lower-level supplemental routines that compute search directions in
> various ways (mostly as "steepest descent").  They are 2D specific and
> require a custom-made FEM solver for simple elliptic equations.
> 
> Of course, the code is general enough to handle different actual
> problems.  (I'm using it both for PDE-constrained shape optimisation and
> for a special image segmentation method.)
> 
> I'm currently thinking about the best ways for publishing these.  Of
> course, it is ultimately for me to decide about that (particularly as
> the maintainer of the level-set package), but I would like to get
> community feedback if possible.  I see the following variants:
> 
> a) Include at least 1) and 2) as new parts in the level-set package.
> IMHO, these parts are general enough to loosely fit into the existing
> package.  On the other hand, it might be better to fully keep out any
> "optimisation" related stuff from the level-set package and focus it
> only on basic geometrical routines with level-sets.
> 
> b) Ask for inclusion of the general parts into the "optim" package.
> This seems an even worse fit and not like the best idea.
> 
> c) Create a new package for shape optimisation with my code, which
> depends on level-set.  Seems like a good idea.  However, there exists a
> vast range of methods employed in shape optimisation.  So I'm not sure
> if it is "ok" to publish a "shape optimisation" package which basically
> just employs the particular methods I use and develop.
> 
> d) Publish a specific package for shape optimisation with the level-set
> method.  Would that still be of interest to Octave Forge?  If not, then
> I can, of course, also publish it separately (but still as free software).
> 
> What do you think?  Any feedback is very welcome!  I would like to make
> my code of the most use to the research community and the users of Octave.
> 
> Yours,
> Daniel
> 
> -- 
> http://www.domob.eu/
> OpenPGP: 1142 850E 6DFF 65BA 63D6  88A8 B249 2AC4 A733 0737
> Namecoin: id/domob -> https://nameid.org/?name=domob
> --
> Done:  Arc-Bar-Cav-Hea-Kni-Ran-Rog-Sam-Tou-Val-Wiz
> To go: Mon-Pri
> 



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

signature.asc
Description: Digital signature