[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: lsqnonlin and nonlin_residmin
From: |
Olaf Till |
Subject: |
Re: lsqnonlin and nonlin_residmin |
Date: |
Sun, 31 May 2015 09:43:14 +0200 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
On Fri, May 29, 2015 at 06:39:26PM +0200, Olaf Till wrote:
> On Thu, May 28, 2015 at 12:51:58PM -0700, AsmaA wrote:
> > 2- Settings. Do we need a matlab compatible function 'optimoptions'?
>
> Probably yes. How this should be done still requires some
> investigation (or maybe you have already thought about this ...).
Would be a nice addition. Since you strive for Matlab compatibility,
it should be a potential task for you.
I'd guess it is much work, and it was not foreseen. Maybe it's better
not to tackle it until everything else on your list is ready. Still,
we do have optimset and it works.
Seems the difference to optimset, as seen by the user, is mainly that
the options object is _displayed_ with additional information specific
to solver and algorithm. But I'm not sure, some things should be
tested in Matlab, e.g.:
1) Is there an error or warning if optimoptions is used to
assign an option not contained in the specified solver? ...
2) ... or to assign an option that does not exist in any solver?
3) Same as (1) or (2), but if option is specified in dot notation?
4) If options are transfered to a different solver with optimoptions,
are there errors or warnings if the new solver does not have some
of these options?
5) Can options returned by optimoptions be used with optimset, and
vice versa?
I haven't found if there is something comparable to 'optimget', but
used with the options object of 'optimoptions' ... maybe there is no
such user function ... (?)
The work splits into two tasks:
1) Make solvers return information specific to each of their
algorithms. This task would be for me, probably. I'd suggest just
to return each algorithms default options in a structure on
request. I'd probably not tackle this until a good plan is ready
for 2).
2) Implement the function 'optimoptions'. Probably chiefly your
task. Some hints:
2.1) Thoroughly check the behaviour in Matlab before, e.g the issues
1)-5) above.
2.2) First make a plan and discuss what your code will do in detail,
please.
2.3) Take care of case-insensitivity of options, but displaying options
in the nominal 'CamelCase'. See how optimset does it.
2.4) You'll need to make a class for the options objects, with methods
for displaying and assigning, the latter also in dot notation.
2.5) 'optimget' should, if possible, work with the new options object
as before with the options of 'optimset', since existing solvers
use 'optimget' to access options. (This should, if possible, be
achieved by the design of the new options class, not by changing
'optimget'.)
Olaf
--
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net
signature.asc
Description: Digital signature