MIQP - also of interest for Rmetrics Portfolio Optimization ...

Diethelm Wuertz wuertz at itp.phys.ethz.ch
Wed Oct 3 00:50:05 CEST 2007


Kurt Hornik wrote:
>>>>>> Brian G Peterson writes:
>>>>>>             
>
>   
>> Kurt Hornik wrote:
>>     
>>> Diethelm/Brian,
>>>
>>> Are you aware of decent open source code for mixed integer quadratic
>>> programming?
>>>       
>
>   
>> I assume that you already know about the GNU Linear Programming Kit?
>>     
>
>   
>> http://www.gnu.org/software/glpk/
>>     
>
>   
>> I've wanted an R interface to this for some time, and considered using 
>> SWIG to generate the first cut of bindings at my previous job.
>>     
>
> Yes.  There is in fact a low-level R API in package glpk, and we are
> working on a high level solver (maybe sparse matrix capable).  But it
> does not seem to have quadratic solvers.
>
>   
>> I don't think they have released quadratic solvers, but I know it's 
>> something they've done some work on, so contacting the project is 
>> probably worthwhile.
>>     
>
> Right.  Or/and maybe also contact the lpSolve people (we currently tend
> to use this solver, and the project seems to be more active in a way
> [glpk may be a one-person show].  We are also playing with COIN-OR but
> again this has no dedicated MIQP solver, it seems.
>
> This would be a good project for the Rmetrics foundation.
>   
I agree ...
DW
> Best
> -k
>
>
>
>
>   
>> There's some Fortran work I found at Argon here:
>>     
>
>   
>> http://www-unix.mcs.anl.gov/~leyffer/solvers.html
>>     
non public ?!
DW
>
>   
>> Although it looks like they've removed their code from the site, so 
>> you'd need to contact them regarding licensing.
>>     
>
>   
>> Sorry I can't be more help here.
>>     
>
>   
>> Regards,
>>     
>
>   
>>    - Brian
>>     
>
>   
Good Open Source solvers are very limited!
Have als a look on Ladislav Luksan's web page --- GNU licensed routines
http://www.cs.cas.cz/~luksan/subroutines.html
DW

Or:
Rdonlp2 (non-public) ---- http://arumat.net/Rdonlp2/
DW



More information about the Rmetrics-core mailing list