[Coin-lpsolver] Shouldn't primalPivotResult update reduced costs?

John J Forrest jjforre at us.ibm.com
Fri Jun 9 04:34:11 EDT 2006


Paulo,

I don't see why the time should have increased compared to 
setObjectiveAndRefresh as I was just doing part of that code.

However I was being stupid.  I had put in the dual update but in an odd 
place in the sequence of the code AND with a bug.  So I have taken out the 
extra parameter I put in yesterday and it should be faster.

John Forrest



"Paulo J. S. Silva" <pjssilva at ime.usp.br> 
Sent by: coin-lpsolver-bounces at list.coin-or.org
06/08/2006 03:45 PM

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

Subject
Re: [Coin-lpsolver] Shouldn't primalPivotResult update reduced costs?







> primalPivotResult is just the last part of a primal iteration in Clp
> (as no other solver implements it).  Normally the reduced costs are
> updated at the beginning of the next iteration.  However I agree that
> if you are using primalPivotResult you would expect the reduced costs
> to be updated.  I have added an extra parameter to primalPivotResult
> to update dual solution which defaults to false so that nobody will be
> affected.  Tell me if it does not do what you want? 

It is working now, thanks. However, the time for a primalPivotResult
increased a lot (it was multiplied by 2). I'll e-mail you personally on
this subject as the discussion may not be relevant for everybody.

> I have always been a little uneasy about the "build your own simplex"
> parts of OsiSolverInterface and assuming that only Clp will ever
> implement it, it may be better to approach it in a different way.  If
> I knew how people use this interface I could add a virtual class to
> Clp - rather like ClpEventHandler which would modify the behavior of
> Clp without continually entering and leaving it.  This would be more
> robust and faster. 

IMHO, if only Clp is going to implement this part of the interface, you
are probably be right. I wrote my code using Osi, instead of using pure
Clp, with the hope to test some day the "same" code with other solver. 
I'll take a closer look at ClpEventHandler and pure Clp. I was only
using it through OsiInterface until now.

Best,

Paulo

_______________________________________________
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/20060609/4c1c2481/attachment.html>


More information about the Clp mailing list