[SCIP] Master problem with barrier
Bahareh Eghtesadi
b.eghtesadi at gmail.com
Tue Oct 27 22:35:44 CET 2015
Dear Matthias,
Thanks for your reply.
Setting it to 'c' still gives the error( assertion !lpCutoffDisabled(set)
failed). Is there a way to change the objective limit? because in the lp.c
file, it comes to this assertion as SCIPlpiIsObjlimExc(lp->lpi) becomes
true. I found the following values when this error happens:
set->lp_disablecutoff = 2, set->nactivepricers =1, and lp->lpiuobjlim =
1e+20
So I am not sure how to prevent this happening. I appreciate any ideas.
Thanks.
On Fri, Oct 23, 2015 at 8:24 AM, Matthias Miltenberger <miltenberger at zib.de>
wrote:
> Dear Bahareh,
>
> it seems that on your problem the barrier does not converge to a solution
> that is both primal and dual feasible.
> Have you tried setting the initialalgo and the resolvealgo to "c"? This
> will perform a simplex-like crossover to generate a basic solution based on
> what the barrier computed. This could help to get a reliable solution.
>
> all the best
> Matthias
>
>
> On 22.10.2015 20:30, Bahareh Eghtesadi wrote:
>
> Dear all,
>
> I have a branch and price algorithm. I am facing a problem when I use the
> barrier method to solve the master problem. I was hoping that you can help
> me.
> When I set both "lp/initalgorithm" and "lp/resolvealgorithm" to barrier
> method, in most cases It stops with : (I have copied a little longer
> version of the lof file at the end of the email)
>
> Barrier limit on dual objective exceeded.
> Infeasible barrier solution (dependent on objective limit).
>
> src/scip/lp.c:14237: lpSolve: Assertion `!(set->lp_disablecutoff == 1 ||
> (set->nactivepricers > 0 && set->lp_disablecutoff == 2))' failed.
>
> I could not find a parameter such as barrier objective limit. I used
> SCIPsetObjlimit to set a higher objective limit ( I am not sure if the
> limit this function is setting is the same as the barrier objective limit)
> . This time it gives the following line as well:
> src/scip/lp.c:14237: lpSolve: Assertion `!(set->lp_disablecutoff == 1 ||
> (set->nactivepricers > 0 && set->lp_disablecutoff == 2))' failed.
>
> For a few instances, setting only "lp/initalgorithm" to barrier and
> leaving "lp/resolvealgorithm" to its default value solves the problem. I
> should also mention that this problem does not happen when master problem
> is solved with Simplex.
>
> The log file when "lp/initalgorithm" = 'b' and "lp/resolvealgorithm"= 'b':
> Iteration log . . .
> Iteration: 1 Dual objective = 10.330177
> Tried aggregator 1 time.
> LP Presolve eliminated 42 rows and 244 columns.
> Aggregator did 19 substitutions.
> Reduced LP has 78 rows, 119 columns, and 599 nonzeros.
> Presolve time = 0.00 sec. (0.33 ticks)
> Parallel mode: using up to 5 threads for barrier.
> Number of nonzeros in lower triangle of A*A' = 494
> Using Approximate Minimum Degree ordering
> Total time for automatic ordering = 0.00 sec. (0.03 ticks)
> Summary statistics for Cholesky factor:
> Threads = 5
> Rows in Factor = 78
> Integer space required = 115
> Total non-zeros in factor = 841
> Total FP ops to factor = 10747
> Itn Primal Obj Dual Obj Prim Inf Upper Inf Dual Inf Inf
> Ratio
> 0 1.6080489e+04 -1.4387026e+02 2.85e+02 1.01e+02 2.25e+03
> 1.00e+00
> 1 1.4132721e+04 -3.4691316e+02 2.52e+02 8.88e+01 1.23e+03
> 1.83e-01
> 2 1.4278598e+04 -5.9373199e+02 2.56e+02 9.01e+01 1.30e+03
> 1.20e-01
> 3 9.9158960e+03 -1.7167304e+03 1.94e+02 6.84e+01 8.37e+02
> 4.26e-02
> 4 5.0149903e+02 -9.9732945e+02 2.01e+01 7.07e+00 3.08e+02
> 8.41e-02
> 5 1.3386365e+02 -2.0932657e+02 4.66e+00 1.64e+00 7.00e+01
> 3.80e-01
> 6 3.2369401e+01 -3.3609945e+01 8.79e-01 3.10e-01 1.36e+01
> 2.14e+00
> 7 2.1874403e+01 -1.7917016e+01 5.50e-01 1.94e-01 7.97e+00
> 3.46e+00
> 8 1.6189845e+01 7.2562159e+00 2.76e-01 9.73e-02 9.12e-01
> 1.44e+01
> 9 2.8249210e+01 1.6791221e+01 3.35e-01 1.18e-01 1.09e+00
> 6.18e+00
> 10 4.5434437e+01 5.6716247e+01 3.38e-01 1.19e-01 1.11e+00
> 3.23e+00
> 11 4.9392601e+01 6.6979256e+01 3.20e-01 1.13e-01 9.47e-01
> 1.53e+00
> 12 5.6458356e+01 9.9330286e+01 3.09e-01 1.09e-01 8.07e-01
> 5.59e-01
> 13 2.2191704e+02 1.0781431e+02 3.05e-01 1.07e-01 8.07e-01
> 2.03e-02
> 14 8.6508663e+01 1.0162578e+02 2.74e-01 9.64e-02 7.82e-01
> 6.30e-03
> 15 3.3035441e+02 7.4121836e+05 2.74e-01 9.66e-02 8.71e-01
> 8.24e-05
> 16 2.5380932e+02 1.7310730e+08 2.99e-01 1.06e-01 5.31e+00
> 2.59e-08
> 17 1.7742307e+02 3.5154843e+11 2.75e-01 9.69e-02 3.96e+04
> 3.27e-12
> 18 1.6774322e+02 3.0499215e+15 2.72e-01 9.59e-02 4.97e+08
> 3.28e-16
> 19 1.6803927e+02 2.9430527e+19 2.61e-01 9.19e-02 1.17e+13
> 3.40e-20
> 20 1.6131741e+02 2.7393729e+23 2.57e-01 8.61e-02 5.92e+16
> 4.39e-24
> * 1.6803927e+02 2.9430527e+19 2.61e-01 9.19e-02 1.17e+13
> 3.40e-20
> Barrier limit on dual objective exceeded.
> Infeasible barrier solution (dependent on objective limit).
> Barrier time = 0.11 sec. (1.56 ticks)
>
> Total time on 5 threads = 0.11 sec. (1.56 ticks)
> vaccine: src/scip/lp.c:14237: lpSolve: Assertion `!(set->lp_disablecutoff
> == 1 || (set->nactivepricers > 0 && set->lp_disablecutoff == 2))' failed.
>
>
>
> Thanks.
>
>
>
> _______________________________________________
> Scip mailing listScip at zib.dehttp://listserv.zib.de/mailman/listinfo/scip
>
>
> --
> \__________________
>
> Matthias Miltenberger
> Zuse Institute Berlin
> Takustr. 7, 14195 Berlinwww.zib.de/miltenbergermiltenberger at zib.de+49 (30) 841 85-245
>
>
--
Regards
Bahareh Eghtesadi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20151027/065ff495/attachment.html>
More information about the Scip
mailing list