[Coin-discuss] SBB improvements

[Jean-Sebastien Roy]

John J Forrest wrote:

> I may have time to get back to SBB.  Are there any requests for 
> improvements in any of the following areas?
>
> 1) General improvements
> 2) New heuristics to obtain solutions
> 3) New cuts (well implementation of known cuts not at present in Cgl)
> 4) New types of variables/sets e.g. special Ordered Sets of type 2. 
>  For instance I have already added a lotsizing variable for a 
> particular project.

My vote is on general improvements (speed and branching heuristics) and 
new heuristics to obtain solutions.

It would be nice to beat GLPK for example. On some generation planning 
problems my colleagues and I work on (small ones, with a few hundred 
variables and constraints, and about a hundred binary variables; one run 
of the software requires solving about a million of them), GLPK 
consistently beat SBB by a factor 2 or 3 (and without using any cuts, or 
an integer presolve, or rounding heuristics).

Regards,

js