[Scip] Assertion failed: branchcand->pseudocands[var->pseudocandindex] == var, file src\scip\branch.c, line 928
Stefan Heinz
heinz at zib.de
Fri Apr 22 13:51:45 MEST 2011
Hi Sebastian,
yes there is bug when you set the node limit to one. We are currently
looking into it. We will come back to you.
Stefan
> Ok, so the same problem happens in your examples.
> Add SCIP_CALL( SCIPsetLongintParam(scip,"limits/nodes", 1) ); before you
> call SCIPsolve
> and
> SCIP_CALL(SCIPprintStatistics(scip, NULL)); in the scip_redcost function
> of the Binpacking or VRP example and you'll get the invalid LP problem
> after the first pricing iteration.
>
> Cheers,
> Sebastian
>
> On 19/04/2011 1:28 PM, Sebastian Ruther wrote:
>> Actually, the assert happens not only on Windows but also on Linux if
>> I use OPT=dbg. Below I used OPT=noblkmem.
>> Again the assertion happens only if I add
>> SCIP_CALL( SCIPprintBestSol(scip, file, FALSE) );
>> or
>> SCIP_CALL( SCIPprintTransProblem(scip, file, NULL, FALSE) );
>> or both. Otherwise I get the 'LP of current node is invalid' followed
>> by 'node limit reached'
>>
>> Sebastian
>>
>> On 19/04/2011 12:09 AM, Sebastian Ruther wrote:
>>> Hello again,
>>>
>>> so I got it all running in Linux and used Valgrind. It comes up with
>>> some "Conditional jump or move depends on uninitialised values" and
>>> has lot's of lost memory, which is expected as I haven't coded
>>> anything that frees memory yet. See attached file.
>>> However, memory leak shouldn't be the issue.
>>> So that leads me back to my earlier question: Why is the LP invalid?
>>> I investigated some more when this problem occurs and got some
>>> strange behavior. But first let me repeat what I'm actually doing.
>>>
>>> I'm implementing a branch and price algorithm and so far I
>>> implemented a reader and a pricer, but (at least in this version) no
>>> branching rule or constraint handler (for the branching). I do have
>>> multiple subproblems that all get called in one iteration of a
>>> pricer. However, in the following I deactivate all but one subproblem.
>>> I load the default plugins and then for the master problem
>>> SCIP_CALL( SCIPsetIntParam(scip,"presolving/maxrestarts",0) );
>>> SCIP_CALL( SCIPsetIntParam(scip, "presolving/maxrounds", 0) );
>>> SCIP_CALL( SCIPsetIntParam(scip, "propagating/maxrounds", 0) );
>>> SCIP_CALL( SCIPsetIntParam(scip, "propagating/maxroundsroot", 0)
>>> );
>>> And for the subproblem
>>> SCIP_CALL( SCIPsetIntParam(subscip, "presolving/maxrounds",
>>> 0) ); disabling the presolver is quicker.
>>> SCIP_CALL( SCIPsetIntParam(subscip, "presolving/maxrestarts",
>>> 0) );
>>> SCIP_CALL( SCIPsetRealParam(subscip, "limits/gap", 0.01) );
>>>
>>> First, if I don't inlcude SCIP_CALL(
>>> SCIPsetLongintParam(scip,"limits/nodes", 1) ); then everything is
>>> fine, so from now on this option is always included.
>>>
>>> Then, in the beginning of pricer::scip_redcost I call
>>> SCIP_CALL(SCIPprintStatistics(scip, file)); (if I
>>> deactivate this I don't have any problems)
>>>
>>> And this causes problems. After the pricing round but before I enter
>>> scip again I get:
>>> [src/scip/solve.c:1341] Warning: pricing has been interrupted -- LP
>>> of current node is invalid
>>> 6.7s| 1 | 0 | 2 | - | 0 | 0 | - | 152 | 259 |
>>> 152 | 259 | 0 | 0 | 0 | -- | 1.270000e+04 | Inf
>>> 6.8s| 1 | 2 | 2 | - | 0 | 0 | 0 | 152 | 259 |
>>> 152 | 259 | 0 | 0 | 0 | -- | 1.270000e+04 | Inf
>>>
>>> SCIP Status : solving was interrupted [node limit reached]
>>> Solving Time (sec) : 6.78
>>> Solving Nodes : 1
>>> Primal Bound : +1.27000000000000e+04 (4 solutions)
>>> Dual Bound : -1.00000000000000e+20
>>> Gap : infinite
>>>
>>> this is on a linux machine. I attached the valgrind output.
>>>
>>> On a windows machine I get the same problem. However, if I add either
>>> SCIP_CALL( SCIPprintBestSol(scip, file, FALSE) );
>>> or
>>> SCIP_CALL( SCIPprintTransProblem(scip, file, NULL, FALSE) );
>>> or both I now get an assert:
>>>
>>> time | node | left |LP iter|LP it/n| mem |mdpt |frac |vars |cons
>>> |cols |rows |cuts |confs|strbr| dualbound | primalbound | gap
>>> t 2.9s| 1 | 0 | 6 | - | 652k| 0 | 0 | 159 | 259 |
>>> 159 | 259 | 0 | 0 | 0 | -- |1.500000e+004 | Inf
>>> [src\scip\solve.c:1312] Warning: pricing has been interrupted -- LP of
>>> current node is invalid
>>> 2.9s| 1 | 0 | 6 | - | 652k| 0 | 0 | 159 | 259 |
>>> 159 | 259 | 0 | 0 | 0 | -- |1.500000e+004 | Inf
>>> * 2.9s| 1 | 0 | 6 | - | 654k| 0 | - | 159 | 259 |
>>> 159 | 259 | 0 | 0 | 0 | -- |1.270000e+004 | Inf
>>> Assertion failed: SCIPlpGetSolstat(lp) == SCIP_LPSOLSTAT_OPTIMAL ||
>>> SCIPlpGetSolstat(lp) == SCIP_LPSOLSTAT_UNBOUNDEDRAY, file
>>> src\scip\branch.c, line 200
>>>
>>> The call stack is
>>>
>>>> msvcr100d.dll!_NMSG_WRITE(int rterrnum=10) Line 217
>>> msvcr100d.dll!abort() Line 71
>>> msvcr100d.dll!_wassert(const wchar_t * expr=0x000000014021a4b0,
>>> const wchar_t * filename=0x000000014021a488, unsigned int
>>> lineno=200) Line 153
>>>
>>> binscip.exe!branchcandCalcLPCands(std::basic_ostream<char,std::char_traits<char>
>>> > & (char) branchcand=0x000000000013f010) Line 200 + 0x3e bytes
>>> binscip.exe!SCIPbranchcandGetLPCands() Line 344 + 0x19 bytes
>>> binscip.exe!solveNode() Line 2849 + 0x56 bytes
>>> binscip.exe!SCIPsolveCIP() Line 3264 + 0x101 bytes
>>> binscip.exe!SCIPsolve() Line 6827 + 0x11f bytes
>>> binscip.exe!runSCIP(int argc=1, char * *
>>> argv=0x00000000004776f0) Line 181 + 0xa bytes
>>> binscip.exe!main(int argc=1, char * * argv=0x00000000004776f0)
>>> Line 238 + 0xe bytes
>>> binscip.exe!__tmainCRTStartup() Line 555 + 0x19 bytes
>>> binscip.exe!mainCRTStartup() Line 371
>>> kernel32.dll!0000000076e3f56d()
>>> [Frames below may be incorrect and/or missing, no symbols loaded
>>> for kernel32.dll]
>>> ntdll.dll!0000000076f72cc1()
>>>
>>> so none of the calls I mentioned above are used.
>>>
>>> Any help is appreciated.
>>>
>>> Sebastian
>>>
>>>
>>>
>>>
>>>
>>>
>>> On 7/04/2011 7:42 PM, Sebastian Ruther wrote:
>>>> Hello,
>>>>
>>>> I went back to an early version of my code to see if this indeed has
>>>> something to do with the branching rule and the constraint handler. So
>>>> this version does not include them at all, instead uses the default
>>>> plugins. When I set the node limit to 1 I get the following error
>>>> after
>>>> the first round of pricing. Note it should have executed 110
>>>> iterations
>>>> in the root node. So this is the output for the master problem after
>>>> iteration 1:
>>>>
>>>> time | node | left |LP iter|LP it/n| mem |mdpt |frac |vars |cons
>>>> |cols |rows |cuts |confs|strbr| dualbound | primalbound | gap
>>>> t 2.9s| 1 | 0 | 6 | - | 652k| 0 | 0 | 159 | 259 |
>>>> 159 | 259 | 0 | 0 | 0 | -- |1.500000e+004 | Inf
>>>> [src\scip\solve.c:1312] Warning: pricing has been interrupted -- LP of
>>>> current node is invalid
>>>> 2.9s| 1 | 0 | 6 | - | 652k| 0 | 0 | 159 | 259
>>>> |
>>>> 159 | 259 | 0 | 0 | 0 | -- |1.500000e+004 | Inf
>>>> * 2.9s| 1 | 0 | 6 | - | 654k| 0 | - | 159 | 259 |
>>>> 159 | 259 | 0 | 0 | 0 | -- |1.270000e+004 | Inf
>>>> Assertion failed: SCIPlpGetSolstat(lp) == SCIP_LPSOLSTAT_OPTIMAL ||
>>>> SCIPlpGetSolstat(lp) == SCIP_LPSOLSTAT_UNBOUNDEDRAY, file
>>>> src\scip\branch.c, line 200
>>>>
>>>> I don't know if this is a different problem or if whatever causes the
>>>> other problem causes this one as well. Why is the LP invalid? I don't
>>>> have Valgrind as I run Windows 7. Is there anything else I can do
>>>> before
>>>> trying to set up something for Valgrind? What other information do
>>>> you need?
>>>>
>>>> Thanks
>>>> Sebastian
>>>>
>>>> On 6/04/2011 10:57 PM, Stefan Vigerske wrote:
>>>>> Hi,
>>>>>
>>>>>> The call stack reads
>>>>>> SCIPsolve
>>>>>> SCIPsolveCIP scip.c line 3229
>>>>>> SCIPnodeFocus(...) solve.c line 3229
>>>>>> treeSwitchPath(tree, blkmem, set, stat, prob, primal, lp,
>>>>>> branchcand, conflict, eventfilter, eventqueue, fork, *node,
>>>>>> cutoff) tree.c line 3984
>>>>>> SCIPeventqueueProcess(eventqueue, blkmem, set, primal, lp,
>>>>>> branchcand, eventfilter) tree.c line 2768
>>>>>> SCIPeventProcess(event, set, primal, lp, branchcand, eventfilter)
>>>>>> event.c line 2174
>>>>>> SCIPbranchcandUpdateVar(branchcand, set, var) event.c line 1390
>>>>>> SCIPbranchcandRemoveVar(branchcand, var) branch.c line 1053
>>>>>> branchcandRemovePseudoCand(branchcand, var) branch.c line 1012
>>>>>> assert(branchcand->pseudocands[branchcand->npseudocands-1] !=
>>>>>> NULL); branch.c line 928
>>>>>>
>>>>>> I've isolated the problem somewhat. However, whether these actions
>>>>>> cause or just trigger detection of the problem I don't know:
>>>>>> The way I implemented the branching is that the scip_activate of
>>>>>> the constraint handler removes some arcs (i.e. variables) from the
>>>>>> sub problem. For this it sets the upper bound to 0 by calling
>>>>>> SCIPchgVarUbNode. I got an error here before saying ERROR:
>>>>>> original variable has no transformed variable attached. I guess
>>>>>> that's because no transformed sub problem exists at this point.
>>>>> If there is no transformed problem, then SCIPchgVarUbNode is not the
>>>>> function to call. Probably you want to use SCIPchgVarUb.
>>>>>
>>>>> Also try running your code through valgrind.
>>>>>
>>>>> Stefan
>>>>>
>>>>>> The pricing problems are generated in the Objpricer::init_sol and
>>>>>> I only update the cost coefficients in the scip_redcost function.
>>>>>> What happens is that the scip_redcost destroys the transformed
>>>>>> problem after it's finished. So every time scip_redcost is
>>>>>> executed it needs to transform the original problem. Because the
>>>>>> constraint handler is called before scip_redcost I call
>>>>>> SCIPtransformProb in the constraint handler. Probably not the most
>>>>>> elegant implementation but apart from that, is there any problem
>>>>>> with calling it here?
>>>>>>
>>>>>> Cheers,
>>>>>> Sebastian
>>>>>> ________________________________________
>>>>>> From: Stefan Vigerske [stefan at math.hu-berlin.de]
>>>>>> Sent: 04 April 2011 21:10
>>>>>> To: Sebastian Ruther
>>>>>> Cc: scip at zib.de
>>>>>> Subject: Re: [Scip] Assertion failed:
>>>>>> branchcand->pseudocands[var->pseudocandindex] == var, file
>>>>>> src\scip\branch.c, line 928
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>>> Hello,
>>>>>>>
>>>>>>> I'm currently implementing a follow-on branching rule and I get the
>>>>>>> following error:
>>>>>>> Assertion failed: branchcand->pseudocands[var->pseudocandindex]
>>>>>>> == var,
>>>>>>> file src\scip\branch.c, line 928
>>>>>>> What does this error mean? What are these pseudocands?
>>>>>> Die pseudocands are branching candidates for pseudo branching, in
>>>>>> principle the binary or integer variables that are not fixed in the
>>>>>> current node.
>>>>>> If you get this assertion, then I guess some datastructures have
>>>>>> been
>>>>>> corrupted.
>>>>>>
>>>>>>> The error occurs after i execute the ObjConshdlr::scip_activate.
>>>>>>> I've implemented the ObjBranchrule::scip_execlp. I have not
>>>>>>> imlemented
>>>>>>> the ObjBranchrule::scip_execps but it looks like it is not
>>>>>>> entered. I
>>>>>>> think the error is because of something in the constraint handler I
>>>>>>> implemented because when I comment out those calls I don't get
>>>>>>> any errors.
>>>>>> To give some helpful comments, it would be useful to get more
>>>>>> information about what you are doing and when this problem exactly
>>>>>> happens.
>>>>>> Can you provide a backtrace from running in a debugger?
>>>>>> Which SCIP functions does your code call?
>>>>>>
>>>>>> Stefan
>>>>>>
>>>>>>> Thanks
>>>>>>> Sebastian
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Scip mailing list
>>>>>>> Scip at zib.de
>>>>>>> http://listserv.zib.de/mailman/listinfo/scip
>>>>>>>
>>>>>> --
>>>>>> Stefan Vigerske
>>>>>> Humboldt University Berlin, Numerical Mathematics
>>>>>> http://www.math.hu-berlin.de/~stefan
>>>>>>
>>>>> .
>>>>>
>>>> _______________________________________________
>>>> 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
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip
>
More information about the Scip
mailing list