[Coin-discuss] Global cuts

[Robin Lougee-Heimer]

Hi Laci:

>I can very well imagine someone wanting to generate cuts based on only
>a subset of the currently existing constraints

Absolutely, this happens all the time. (Actually there's more to it,
becuase you want to be able to pass around substructures, like SOS sets and
the like so you don't have to constantly re-derive that information, and
also manage changes to the substructure when you do iterative algorithms
using probing, and the like.). I'd have to check, but I think the way this
was done was to include the range of row indices to be examined as an
argument to the construct of the specific cutGenerator. Of course, another
way would be to simply add an argument with the rows-to-be-examined
information to the generateCuts method.

>In that case
>creating a SolverInterface object just to pass in the data (since no
>LP will be solved by that solverinterface) can be really harmful to
>the performance.

Why does wanting to examine a subset of constraints necessitate creating a
new SolverInterface object?


Robin


----------------------------------------------------------------------------------

Robin Lougee-Heimer
IBM TJ Watson Research Center
ph: 914-945-3032   fax: 914-945-3434
robinlh at us.ibm.com
http://www.coin-or.org



ladanyi at watson.ibm.com@www-124.southbury.usf.ibm.com on 01/29/2003 10:50:23
PM

Please respond to coin-discuss at www-124.southbury.usf.ibm.com

Sent by:    coin-discuss-admin at www-124.southbury.usf.ibm.com


To:    coin-discuss at www-124.southbury.usf.ibm.com
cc:
Subject:    Re: [Coin-discuss] Global cuts



I can very well imagine someone wanting to generate cuts based on only
a subset of the currently existing constraints (especially if cut
generation takes place elsewhere than solving LPs does). In that case
creating a SolverInterface object just to pass in the data (since no
LP will be solved by that solverinterface) can be really harmful to
the performance. Not just the copying (that would not be a big deal,
after all, most likely the matrix is going to be scanned several time,
so one extra scan does not matter), but the extra allocation of memory
needed to store the extra copy of the matrix may force paging, which
is bad.

--Laci

On Wed, 29 Jan 2003, Robin Lougee-Heimer wrote:

> Laci:
>
> >Basically I'd like to
> >make it possible to pass in the problem description and solution in some
> other
> >way than passing in a solverinterface.
>
>  What's the benefit/motivation for  passing in the packed matrix and
primal
> solution as oppose to a reference to the solver interface?
>
> Robin
>
>
>
----------------------------------------------------------------------------------

>
> Robin Lougee-Heimer
> IBM TJ Watson Research Center
> ph: 914-945-3032   fax: 914-945-3434
> robinlh at us.ibm.com
> http://www.coin-or.org
>
>
>
> ladanyi at watson.ibm.com@www-124.southbury.usf.ibm.com on 01/29/2003
02:52:12
> PM
>
> Please respond to coin-discuss at www-124.southbury.usf.ibm.com
>
> Sent by:    coin-discuss-admin at www-124.southbury.usf.ibm.com
>
>
> To:    coin-discuss at www-124.southbury.usf.ibm.com
> cc:
> Subject:    Re: [Coin-discuss] Global cuts
>
>
>
> Sounds like a good idea...
>
> While we are talking about ideas: I propose to add a couple of more
> generateCuts() methods to the cut generator base class. Basically I'd
like
> to
> make it possible to pass in the problem description and solution in some
> other
> way than passing in a solverinterface. Like passing in a packed matrix
and
> a
> pointer to the primal solution, or passing in a set of pointers (i.e., a
> C-style interface).
>
> Any objections?
>
> --Laci
>
> On Tue, 28 Jan 2003, John J Forrest wrote:
>
> >
> > I wish to add a bool flag to OsiCut to say if it has global validity.
> >
> > Any objections?
> >
> > John Forrest
> >
> > _______________________________________________
> > Coin-discuss mailing list
> > Coin-discuss at www-124.ibm.com
> > http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-discuss
> >
>
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at www-124.ibm.com
>  http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-discuss
>
>
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at www-124.ibm.com
> http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-discuss
>

_______________________________________________
Coin-discuss mailing list
Coin-discuss at www-124.ibm.com
 http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-discuss