[Coin-lpsolver] Best way to bring an instance into CLP?
Lasse Kliemann
lasse-coin-2008 at plastictree.net
Fri Jan 18 08:08:34 EST 2008
* Message by -Lasse Kliemann- from Fri 2008-01-18:
> There are so many ways to bring an LP into the CLP solver.
>
> I figured out at least the following:
>
>
> 1) CoinPackedMatrix --> ClpModel (via loadProblem)
>
> 2) CoinModel --> ClpModel (via loadProblem)
>
> 3) CoinBuild --> CoinModel (via addColumns or addRows)
> --> ClpModel (via loadProblem)
>
> 4) CoinBuild --> ClpModel (via addColumns or addRows)
>
> 5) CoinPackedMatrix --> OsiClpSolverInterface (via loadProblem)
>
> 6) CoinModel --> OsiClpSolverInterface (via loadFromCoinModel)
>
> 7) CoinBuild --> CoinModel (via addColumns or addRows)
> --> OsiClpSolverInterface (via loadFromCoinModel)
Oh, there are even more:
9) CoinBuild --> OsiClpSolverInterface (via addRows or addCols)
And we can also load rows or columns given by vectors of indices and elements
(what I assumed to be the starting point all the time) directly into ClpModel
or OsiClpSolverInterface via addRow or addCol.
> Which of these is the best in terms of memory usage?
> Which is the best in terms of runtime efficiency?
>
> As far as I understood, CoinBuild helps to boost runtime efficiency. Hence,
> number 3 or number 7 could be the fastest solution. However, memory usage is
> an issue in my application.
>
> So far, I've been using alternative number 6, which I found pretty fast
> already (even without the use of CoinBuild). The program aborts with 'out of
> memory' on instances of a certain size. These instances can be solved,
> however, when loaded directly into GLPK; I have not tried loading them
> directly into CLP yet. Before I try all possible ways to find out the best,
> maybe someone can give me a hint.
>
> Thanks a lot for your advice!
>
> Lasse
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 827 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/clp/attachments/20080118/4b4ac4bb/attachment.sig>
More information about the Clp
mailing list