[Os-project-managers] SMPS/Stochastic extensions

[Horand Gassmann]

"R. Kipp Martin" <kmartin at chicagobooth.edu> wrote:

Hi Kipp,

I have taken the liberty of including Jun in this discussion by  
copying to OS-project-managers, which also allows us to archive for  
future use.

> Gus wrote:
>>
>> As far as I know, Smi is the only open-source SMPS parser. There is, of
>> course, the input system for the MSLiP code, which I have been giving away
>> free to researchers, but it's not exactly open-source. If we ever  
>> get serious
>> about the stochastic stuff, we should think hard what would be the best way
>> to get started: tweak and debug Smi, translate SMPSReader into C++, try to
>> figure out the Fortran/C++ interface, or just start from scratch.
>
> Several strategy questions:
>
> 1) Should we devote resources to having OS read SMPS or just parse  
> an OSiL instance representation of a stochastic optimization problem.

I think it is absolutely essential that we have an SMPS parser. There  
are just too many problems in SMPS format out there; if we want people  
to use OS/SE, we will have to allow a growth path, just as we provide  
an MPS reader for deterministic problems. We can make use of  
third-Party code (like Smi or OSI/CoinUtils), but we have to provide  
something. Of course the OSiL and SMPS files should both be processed  
into the same internal data representation.

> 2) I have started to implement a decomposition algorithm in OS. See
>
> OS/applications/columnGen
>
> I tried to make this fairly generic so it can be applied to both  
> Benders and DantzigWolfe kinds of problems. Should we work on making  
> this a stochastic solver also?

Having a decomposition solver along with a deterministic equivalent  
would be a huge plus. And it would obviously be useful if we can  
leverage existing code.

Cheers

gus