[Osi] Solving master problems using interior point method

[Matthew Galati]

This reminds me of an issue I found when I tried this a while back.

CC'ing OSI list.

OSI does not seem to support interior point solvers. For example, there is
no OsiDoInteriorInResolve option, or anything like that.

OSI folks - if I want to invoke an interior solver of Cpx or Clp, for
example, how do I do it?

Matt






On Wed, Jan 5, 2011 at 4:12 AM, Jonas Christoffer Villumsen <jcvi at man.dtu.dk
> wrote:

>  Hi all,
>
>
>
> Does anyone have experience solving master problems with an interior point
> method?
>
>
>
> There is a parameter SolveMasterUpdateAlgo that can be set to 2 (barrier),
> but this doesn’t seem to have an effect.
>
>
>
> In DecompAlgo.cpp I tried to set DO_INTERIOR.  I managed to get it to run,
> but it terminates with an assertion failure – saying that Barrier limit on
> dual objective exceeded.
>
> I have pasted some of the log below (full log attached).  I guess the
> barrier algorithm doesn’t find a feasible solution (within the set limit),
> and the infeasible state is not set properly (as stated in the comments of
> the code). Is this the case? And if so any ideas how to come by that?
>
>
>
> Thanks,
>
> Jonas
>
>
>
> D-ALGO   : 4.35     [CPU: 4.37    ] <--- solutionUpdate() --------
>
> Tried aggregator 1 time.
>
> CPX0000  Tried aggregator 1 time.
>
> LP Presolve eliminated 43 rows and 53 columns.
>
> CPX0000  LP Presolve eliminated 43 rows and 53 columns.
>
> Reduced LP has 36 rows, 44 columns, and 76 nonzeros.
>
> CPX0000  Reduced LP has 36 rows, 44 columns, and 76 nonzeros.
>
> Presolve time =    0.02 sec.
>
> CPX0000  Presolve time =    0.02 sec.
>
> Parallel mode: using up to 2 threads for barrier.
>
> CPX0000  Parallel mode: using up to 2 threads for barrier.
>
> Number of nonzeros in lower triangle of A*A'CPX0000  Number of nonzeros in
>
> lower triangle of
>
>  = 144
>
> CPX0000  Number of nonzeros in lower triangle oCPX0000  A*A'
>
> Using Approximate Minimum Degree ordering
>
> CPX0000  Using Approximate Minimum Degree ordering
>
> Total time for automaticCPX0000  Total time for
>
>  ordering = 0.00 sec.
>
> CPX0000  Total time foCPX0000  automatic
>
> Summary statistics for Cholesky factor:
>
> CPX0000  Summary statistics for Cholesky factor:
>
>   Threads                   = 2
>
> CPX0000    Threads                   = 2
>
>   Rows in Factor            = 36
>
> CPX0000    Rows in Factor            = 36
>
>   Integer space required    = 36
>
> CPX0000    Integer space required    = 36
>
>   Total non-zeros in factor = 180
>
> CPX0000    Total non-zeros in factor = 180
>
>   Total FP ops to factor    = 1140
>
> CPX0000    Total FP ops to factor    = 1140
>
>  Itn      Primal Obj        Dual Obj  Prim Inf Upper Inf  Dual Inf
>
> CPX0000   Itn      Primal Obj        Dual Obj  Prim Inf Upper Inf  Dual Inf
>
>    0  1.2727273e+021  4.0000000e+000 1.75e+021 0.00e+000 1.12e+002
>
> CPX0000     0  1.2727273e+021  4.0000000e+000 1.75e+021 0.00e+000 1.12e+002
>
>    1  5.9609448e+020  5.6913266e+020 3.93e+005 0.00e+000 1.99e+001
>
> CPX0000     1  5.9609448e+020  5.6913266e+020 3.93e+005 0.00e+000 1.99e+001
>
>  Barrier limit on dual objective exceeded.
>
> CPX0000   Barrier limit on dual objective exceeded.
>
>  Infeasible barrier solution (dependent on objective limit).
>
> CPX0000   Infeasible barrier solution (dependent on objective limit).
>
>
>
> Total real time on 2 threads =    0.09 sec.
>
> CPX0000
>
> CPX0000  Total real time on 2 threads =    0.09 sec.
>
> Solution update n_cols:97         n_rows: 79         n_iter: 0
> time:
>
> 0.109
>
> Iteration Count               : 0
>
> isAbandoned()                 : 0
>
> isProvenOptimal()             : 0
>
> isProvenPrimalInfeasible()    : 0
>
> isProvenDualInfeasible()      : 0
>
> isPrimalObjectiveLimitReached : 0
>
> isDualObjectiveLimitReached   : 0
>
> isIterationLimitReached       : 0
>
> Assertion failed: 0, file ..\..\..\..\src\DecompAlgo.cpp, line 2434
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/osi/attachments/20110105/0cdef489/attachment.html>