[Cbc] New trunks and note on infeasibility ray.

[Giacomo Nannicini]

Hello all,
first and foremost: thanks a million for your efforts John.

Second, a small clarification: the paper "On the safety of Gomory cut
generators" is currently *under review* for Mathematical Programming
Computation. At the current stage it is unpublished. As pointed out by
John, the manuscript is available on my webpage.

Giacomo

On Tue, Mar 19, 2013 at 10:19 PM, John Forrest
<john.forrest at fastercoin.com> wrote:
> I have produced some new trunk versions.  I found I had been updating the
> wrong versions so Cbc/trunk will be updated again in a day or two.
>
> Cgl has some fixes plus two cut generators donated by Giacomo Nannicini.
> One is an alternative reduce and split generator (G. Cornuejols and G.
> Nannicini. Practical strategies for generating rank-1 split cuts in
> mixed-integer linear programming. Mathematical Programming Computation,
> 3(4):281-318, 2011) while the other is a robust Gomory generator (G.
> Cornuejols, F. Margot and G. Nannicini. On the safety of Gomory cut
> generators. Mathematical Programming Computation) - see
>
> http://faculty.sutd.edu.sg/~nannicini/index.php?page=publications
>
> As reduce and split cuts can be slow, the full benefit will not be felt
> before my next Cbc/trunk when there will be a parameter for specifying
> number of rounds to do for expensive cuts.
>
> Coinutils/trunk has a few changes to presolve for special variables and
> badly scaled problems.
>
> Cbc will change again, but one feature of interest to Allan Wechsler  is
> -extraVariables 20 (or some such) which switches on a trivial re-formulation
> to group variables with same cost together.
>
> The feature which may affect people is for returning a primal infeasibility
> ray.  The previous code passed the unit test, but on close examination that
> was when Clp had switched to Primal.  Hopefully I have now got the signs
> right for Dual as well.  So if you use infeasibility rays - try trunk and if
> fails reverse signs!  At least it should be more consistent - and should be
> same as other Lp codes.
>
> John Forrest
>
>
>
>
> _______________________________________________
> Cbc mailing list
> Cbc at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cbc
>