[Scip] [Fwd: Re: Problem with debug solution]

[michael.winkler at zib.de]

Please add a "#define SCIP_DEBUG" in the beginning of the prop_probing.c
file and recompile the code. In the output you can identify on which
variable probing detected infeasibility.

Probing tries to derive better bounds by tentatively setting (binary)
variables to certain value (bounds) and then starting a propagation
algorithm. This calls all propagation methods, probably also a propagation
method of your constraint handler.

E.g. x in {0,1}, probing 1. fix x = 0, propagate, 2. fix x = 1 propagate
     if 1. results in an infeasibility x will be fixed to 1
     if 2. results in an infeasibility x will be fixed to 0
     if 1. led to y >= a and in 2. y >= b then the lower bound on y will
be set to y_lb = MIN(a,b)

If you send the probing debug output, please send it to me directly,
instead of using the mailing list.

Best, Michael

> Hi,
>
> I increased the verbosity level.
> Now, SCIP tells me that the probing propagator
> detected the infeasibility. If I switch off probing during
> presolving, this leads to the optimal solution.
>
> The function SCIPdebugCheckConss() has no effect in my constraint
> handler.
> Is there a possibility to print out the reason for the infeasibility
> during probing?
>
> Best, Alex.
>
>
>
> On Tue, 2014-05-06 at 09:23 +0200, Alexander Schnell wrote:
>> Hi Michael,
>>
>> thank you for your suggestions.
>>
>>
>> I don't know exactly how to test the numerical stability
>> of my problem.
>> What I can say is that I am using big-M constraints where I try to work
>> with constants which are as small as possible.
>> Moreover, I work with the product of two binary variables
>> which I model by the and-constraint.
>> I think these two features could lead to numerical problems.
>> However, I tested my approach with 3240 different runs
>> (on 270 instances) and the presolving problem only appears in 0,2% of
>> the runs.
>>
>> Best, Alex.
>>
>>
>> On Mon, 2014-05-05 at 23:31 +0200, michael.winkler at zib.de wrote:
>> > Also for the mailing list...
>> >
>> > ---------------------------- Original Message
>> ----------------------------
>> > Subject: Re: [Scip] Problem with debug solution
>> > From:    bzfwinkm at zib.de
>> > Date:    Mon, May 5, 2014 9:18 pm
>> > To:      "Alexander Schnell" <alexander.schnell at univie.ac.at>
>> > --------------------------------------------------------------------------
>> >
>> > Hi Alexander,
>> >
>> > > Dear SCIP team,
>> > >
>> > > I implemented a constraint handler and there seems
>> > > to be a problem:
>> > >
>> > > With a certain parameter setting, SCIP tells me that an instance is
>> > > infeasible, whereas I know that there must be an optimal solution
>> for
>> > > that instance. SCIP detects the infeasibility during presolving.
>> > > If I switch off presolving, SCIP finds the optimal solution.
>> > >
>> > > When I try to run SCIP with a debug solution (#define
>> > > SCIP_DEBUG_SOLUTION ... in debug.h), it is read
>> > > but SCIP does not print out any violation detected for the debug
>> > > solution. Is this because the infeasibility was detected during
>> > > presolving?
>> > >
>> > > Do you have an advice for the debugging of such situations, i.e.
>> > > a wrong infeasibility detection during presolving?
>> > >
>> >
>> > You can set the verbosity of SCIP to 5 in the interactive shell, this
>> will
>> > hopefully print the plugin that detected the infeasibility.
>> > How numerical stable is your problem?
>> >
>> > In SCIP last release 3.1.0 we added SCIPdebugCheckConss() to the debug
>> > mechanism. This can be call from a constraint handler to maybe check
>> if
>> > some presolving step rendered certain constraints to violate the debug
>> > solution.
>> >
>> > Best, Michael
>> >
>> > > Thank you in advance.
>> > >
>> > > Best regards,
>> > > Alexander Schnell
>> > >
>> > >
>> > >
>> > >
>> > > _______________________________________________
>> > > Scip mailing list
>> > > Scip at zib.de
>> > > http://listserv.zib.de/mailman/listinfo/scip
>> > >
>> >
>> >
>> >
>> > _______________________________________________
>> > Scip mailing list
>> > Scip at zib.de
>> > http://listserv.zib.de/mailman/listinfo/scip
>>
>>
>> _______________________________________________
>> Scip mailing list
>> Scip at zib.de
>> http://listserv.zib.de/mailman/listinfo/scip
>
>
>