[Coin-discuss] Warmstart/basis reuse

[John J Forrest]

Esben,

Presolve tries to update basis as it reduces the size of the problem but 
it does have a much better chance of doing this correctly with the column 
solution as you suggest.  If you start with a basis of full rank then 
every time a row is removed something has to leave basis - but where to? 
If the solution value is 7.0 and the removed row puts an upper bound of 
7.0 on the variable then it is a good guess to put the variable at its 
upper bound.  Without solution values the code has no idea which bound to 
put it to.

John



Esben Mose Hansen <esben at ange.dk> 
Sent by: coin-discuss-bounces at list.coin-or.org
07/18/2007 11:10 AM
Please respond to
Discussions about open source software for Operations Research 
<coin-discuss at list.coin-or.org>


To
Discussions about open source software for Operations Research 
<coin-discuss at list.coin-or.org>
cc

Subject
[Coin-discuss] Warmstart/basis reuse






Hi,

I'm trying to implement warmstart in my solver that uses 
OsiClpSolverInterface::initialSolve() as it's solver. Only, it doesn't 
work 
like I would expect.

These are the steps I use:

1. Solve a problem using OsiClpSolverInterface::initialSolve() 
2. Read the basis with solver.getWarmStart()
3. Load the same problem with (with readMps())
4. Set the basis with solver.setWarmStart()
5. solve the problem again.


In my opinion 5 should be quite fast, but it is in reality slower than the 

original solving, by 30 seconds or so. Any clue to what I am doing in 
wrong? 
I inserted some getBasis()->print() in intialSolve(), and could verify 
that 
the basis on the solved ClpSimplex model was indeed the same as was set on 

the ClpSimplex when solving 5. I even checked that the presolved ClpModel 
(called model model2 in the code) matched.

What have I missed? Does the solver need additional information in step 5, 

such as the column solution from step 1?

Thanks for reading :)

-- 
regards, Esben
_______________________________________________
Coin-discuss mailing list
Coin-discuss at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/coin-discuss

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/coin-discuss/attachments/20070718/5caadd22/attachment.html>