sci.op-research

ltloan@gmail.com wrote:

>
> So far I use the default LP algorithm, which I assume dual simplex. So
> should I try primal simplex at the beginning of the column generation
> subprocess, switch to dual after a cut is found, then switch back to
> primal simplex, and so on?
>

I might be wrong, but I don't think you can switch algorithms on the
fly. You set the NodeAlg parameter at the outset and that's that. So
it's either use dual simplex and live with possibly slow solution of
node LP after adding columns, or use primal simplex and live with
possibly slow solution of the node LP after adding a cut. The only way
to decide which is better is to try both.

Note that the default value for NodeAlg is "automatic", which usually
means "let CPLEX decide". The 10.2 manual says this always reverts to
dual simplex, but indicates that some future version might have CPLEX
deciding on the fly. I don't know if 11.0 is that future version.

/Paul