[SCIP] spx vs spx1

[Matthias Miltenberger]

Dear Stuart,

thanks for sharing your results!

When modifying a central component like the LP solver it is often 
unpredictable what might change with regard to SCIP's performance. There 
are a few differences between the two interfaces (e.g. persistent 
scaling), although they run exactly the same solver under the hood. 
Maybe you could send me some details (problem instances, settings, log 
files) about your findings so that I can have a closer look.

I would also like to address your previous question concerning CLP and 
SoPlex.

While CLP is a very efficient and advanced LP solver as demonstrated in 
the Mittelmann benchmarks, it's important to keep in mind that this is 
comparing pure LP performance. SoPlex is lacking some presolving 
techniques to get a better performance there. Within SCIP though, you 
can see that SoPlex is performing a lot better and is even quite close 
to the speed of SCIP using CPLEX.
There is simply a lot of performance variability involved when dealing 
with such complex codes and my suggestion to you would be to try 
different settings or solvers from time to time - and please let us know 
when SCIP/SoPlex is especially under- or over-performing :-)

Happy SCIPing!
Matthias


On 02-Dec-17 23:58, Stuart Rogers wrote:
> I use SCIP through GAMS, for which soplex2 or soplex can be used as 
> the LP solver: 
> https://www.gams.com/latest/docs/S_SCIP.html#SCIP_OPTIONS. For GAMS 
> SCIP, I believe soplex2 calls the new interface spx, while soplex 
> calls the older interface spx1. In my tests solving a MINLP with GAMS 
> SCIP, I find that SCIP converges more rapidly when using soplex (which 
> uses the older interface) compared to soplex2 (which uses the newer 
> interface). Can you explain this? Thanks, Stuart
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip

-- 
\__________________

Matthias Miltenberger
Zuse Institute Berlin
Takustr. 7, 14195 Berlin
www.zib.de/miltenberger
miltenberger at zib.de
+49 (30) 841 85 245

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20171203/61c95d05/attachment.html>