[Coin-discuss] Access to dual variable corresponding to a newrow

[Laszlo Ladanyi]

Actually, Pierre Pesnau is using BCP with Osi with CPLEX to do branch and cut,
and it works just fine for him. I'm not sure that's sufficient proof, since in
B&C one usually uses the primal, not the dual solution, but I think it
supports what we thought cplex is doing. 

Anyway I think we should pose the question to cplex. Although they may be
reluctant to commit to anything (they might discover some brilliant scheme
that uses a different storage scheme and much more efficient). I guess we just
have to lay down the specifications for Osi and the onus should lay on the
maintainer of a specific interface to do the bookkeeping if necessary.

--Laci

On Wed, 5 Dec 2001, Jonathan Eckstein wrote:

> We can probably just test this.  But if not, it's very bad, because it
> means OSI is not usable for branch and cut when the Solver is CPLEX. 
> That seems like something that ought to work if you want OSI to catch
> on.  If there are problems, then we'd better add some functionality.
> 
> JE
> 
> 
> Laszlo Ladanyi wrote:
> > 
> > I doubt that that would be the case. It's much simpler to compress those
> > columns the deleted row intersected (they'd have to maintain a list of holes
> > separately for each column otherwise -- a nightmare) and append added rows to
> > the end. Still, only cplex can answer the question. Anybody from ILOG on the
> > list? could you give an answer?
> > 
> > --Laci
> > 
> > On Wed, 5 Dec 2001, Jonathan Eckstein wrote:
> > 
> > > I think Mikhail is concerned that in CPLEX, the indices may not just be
> > > added at the end, but may occupy "holes" left by previously deleted
> > > rows. Perhaps that's not the case. Is it always true that when you
> > > delete rows, the later rows get renumbered to fill the resulting gap?
> > >
> > > -- Jonathan
> > >
> > >
> > > Laszlo Ladanyi wrote:
> > > >
> > > > Although it is not explicitly stated in the documentation I have always
> > > > assumed so far that new rows are added to the "bottom" of the matrix getting
> > > > the next so many indices and when rows are deleted the rows are reindexed from
> > > > 0 to the new number of rows. Since it's the user who calls the row
> > > > addition/deletion routines she can easily keep track of the constraints.
> > > >
> > > > I admit it's a bit of a burden, but otherwise the row structure must contain
> > > > an index for each row and what would that index be? The index of the row from
> > > > the beginning of times? The user has to keep track of things in that case,
> > > > too, since probably a number of rows would be permanently thrown out, both
> > > > from the solver and from the user's data structures.
> > > >
> > > > All in all I'd vote for keeping things as they are, since it's not clear that
> > > > Osi can take easily over some work from the user.
> > > >
> > > > --Laci
> > > >
> > > > On Wed, 5 Dec 2001, Mikhail Nediak wrote:
> > > >
> > > > > Hi,
> > > > >
> > > > > I have a question about COIN/OSI.
> > > > >
> > > > > When I add rows to the LP (or cuts for that matter) is there a way in
> > > > > COIN/OSI to trace what indices they will have after subsequent operations
> > > > > on the matrix? This may be important, for example, if one needs values of
> > > > > dual variables for the cuts.
> > > > >
> > > > > CPLEX has function CPXgetrowindex that lets user find row index by its
> > > > > name. I could not, however, find anything of this sort in
> > > > > OsiSolverInterface. Does the user have to trace positions of the rows
> > > > > after matrix modifications by him/herself? It would seem reasonable for
> > > > > the solver to support this functionality.
> > > > >
> > > > > Thanks.
> > > > > Best regards,
> > > > > Mikhail Nediak
> > > > >
> > > > > _______________________________________________
> > > > > 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
> 
> -- 
> 
> Associate Professor Jonathan Eckstein
> MSIS Department, Faculty of Management, Rutgers University
> 
>   TEACHING ADDRESS                 RESEARCH ADDRESS
> +------------------------------+--------------------------------+
> | 255 J.H. Levin Building      |   RUTCOR, Room 148             |
> | 94 Rockafeller Road          |   640 Bartholomew Road         |
> | Livingston Campus            |   Busch Campus                 |
> | Rutgers University           |   Rutgers University           |
> | Piscataway, NJ 08854 USA     |   Piscataway, NJ 08854 USA     |
> | (732) 445-0510               |   (732) 445-3596               |
> | FAX (732) 445-6329           |   FAX (732) 445-5472           |
> +------------------------------+--------------------------------+
> 
> jeckstei at rutcor.rutgers.edu    
> 
> http://rutcor.rutgers.edu/~jeckstei/
> _______________________________________________
> Coin-discuss mailing list
> Coin-discuss at www-124.ibm.com
> http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-discuss
>