[Coin-lpsolver] Parametric Analysis using CLP + OSI

[John J Forrest]

Paulo,

To build a robust parametric solver is difficult as you have to avoid 
degeneracy etc.  It looks as if the code thinks it is making zero moves so 
is slightly perturbing the problem to try and make some progress.  The fix 
you are using will reset these tiny perturbations.  If you don't need 
speed then I would just use plain primal after updating objective and if 
you have a zero move then add a small amount anyway before doing primal.

John Forrest



"Paulo J. S. Silva" <pjssilva at ime.usp.br> 
Sent by: coin-lpsolver-bounces at list.coin-or.org
03/08/2005 01:21 PM

To
coin-lpsolver at list.coin-or.org
cc

Subject
Re: [Coin-lpsolver] Parametric Analysis using CLP + OSI






Ops...

I am still having problems with the parametric analysis code. It looks
like John changes to primalPivotResult still don't refresh everything
needed.

For now my code may face numerical problems if I update the objective
function before primalPivotResult. After many pivots the function is
changing a little the reduced costs (it should remain the same as I said
in the last e-mail). The difference is up to 1.0e-8, but this is enough
for the code to detected incorrectly unboundedness of the problem.

If I move the objective function after the primalPivotResult everything
works. It also work if I leave the objective update before the pivot
function but call

s.disableSimplexInterface()
s.enableSimplexInterface(True)

after the pivot. I believe this two calls refresh some data.

Unfortunately, I have to give a seminar in two days and I still don't
have the slides. So I wont be able to work for two days on this. When I
come back I'll look at the paraobj code John added to the unit tests to
see if I can reproduce the error my reading the MPS file with the
problem description.

Best,

Paulo

-- 
Paulo José da Silva e Silva 
Professor Assistente do Dep. de Ciência da Computação
(Assistant Professor of the Computer Science Dept.)
Universidade de São Paulo - Brazil

e-mail: pjssilva at ime.usp.br        Web: http://www.ime.usp.br/~pjssilva

Teoria é o que não entendemos o    (Theory is something we don't)
suficiente para chamar de prática. (understand well enough to call) 
                                   (practice)



_______________________________________________
Coin-lpsolver mailing list
Coin-lpsolver at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/coin-lpsolver

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/clp/attachments/20050309/4bee6fe0/attachment.html>