[Scip] Problem with Conflict Analysis

[Ambros Gleixner]

Hi Alex,

this is a good remark.  The point is that it is a tighter *global* 
bound.  Hence in your example,  if s has a lower bound of 5, the reasons 
"s >= 3" and "s >= 5" are equivalent: an empty list of bound changes. 
(I.e., t >= 6 should be globally valid.)

Thanks,

Ambros


PS: Maybe you want to confirm for your safety that the tighter value 
from SCIPvarGetLbAtIndex() is the same as SCIPvarGetLbGlobal().



Am 10.10.2013 18:31, schrieb Alexander Schnell:
> Thank you for this information. This helped very much but
> I have another remark.
>
> I agree on the part that the tighter bound may be used
> as a reason for the propagation, but i think the old lower
> bound would lead to a better explanation:
>
> assume that s >= 3 lead to the bound change of t >= 6
> and the tighter global lower bound 5 is introduced for s.
>
> Then the old explanation s>= 3 -> t >= 6 is stronger than
> s>= 5 -> t >= 6.
>
> Alex
>
>
> On Thu, 2013-10-10 at 17:50 +0200, Ambros Gleixner wrote:
>> That means you may use the tighter bound when computing a reason for the
>> propagation.  This is preferable since it can give you a smaller reason
>> than the bound that was known when the propagation was actually
>> performed; smaller meaning a reason with less changes compared to the
>> global bounds now.
>>
>> Hope that helps,
>>
>> ambros
>>
>>
>>
>> Am 10.10.2013 17:44, schrieb Alexander Schnell:
>>> Hi Ambros,
>>>
>>> yes, the lower bound is tighter.
>>>
>>> Alex
>>>
>>>
>>>
>>> On Thu, 2013-10-10 at 17:20 +0200, Ambros Gleixner wrote:
>>>> Hi Alexander,
>>>>
>>>> is the value of the variable responsible for the lower bound change
>>>> tighter than the value that was stored in inferinfo?
>>>>
>>>> I think this could be correct, if global bound changes have occured in
>>>> the meantime.  Can somebody confirm this?
>>>>
>>>> Ambros
>>>>
>>>>
>>>>
>>>>
>>>> Am 10.10.2013 17:15, schrieb Alexander Schnell:
>>>>> Dear Marc,
>>>>>
>>>>> my problem is not about the variable
>>>>> for which the lower bound change is applied (infervar) but about the
>>>>> variable responsible for the lower bound change
>>>>> (i.e. the lower bound of this variable).
>>>>>
>>>>> For this variable SCIPvarGetLbAtIndex(...,...,TRUE) and
>>>>> SCIPvarGetLbAtIndex(...,...,FALSE) are the same values.
>>>>>
>>>>> I integrate the lower bound of this variable as an integer inferinfo
>>>>> in the function SCIPinferVarLbCons(). Moreover, if I query
>>>>> the lower bound of this variable in the CONS_RESPROP function
>>>>> by SCIPvarGetLbAtIndex(), this does not correspond to the
>>>>> inferinfo value.
>>>>>
>>>>> Note, that I can be sure that the value returned
>>>>> from SCIPvarGetLbAtIndex() is integer (I tested it by
>>>>> SCIPisFeasIntegral()).
>>>>>
>>>>> Best regards,
>>>>> Alexander
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> On Thu, 2013-10-10 at 16:09 +0200, Marc Pfetsch wrote:
>>>>>>
>>>>>> Dear Alexander,
>>>>>>
>>>>>> the function SCIPvarGetLbAtIndex() contains a parameter in which you can
>>>>>> specify whether the bound change in questions has already been applied.
>>>>>> To which value have you set this value? In principle, you should be able
>>>>>> to reconstruct the bound change at that time (using infervar and the
>>>>>> above function).
>>>>>>
>>>>>> Please also check whether the bound change via SCIPinferVarLbCons()
>>>>>> actually changed the bound (check return values infeasible and tightened).
>>>>>>
>>>>>> You can look at other examples of conflict resolution, e.g.,
>>>>>> cons_linearordering.c in examples/LOP. Please note that implementing
>>>>>> conflict resolution always is a tricky thing.
>>>>>>
>>>>>> If this does not explain the problem, we would need more details from
>>>>>> you in order to be able to reconstruct the problem. Possibly off this
>>>>>> list ...
>>>>>>
>>>>>> Best
>>>>>>
>>>>>> Marc
>>>>>>
>>>>>>
>>>>>> On 10.10.2013 15:32, Alexander Schnell wrote:
>>>>>>> Dear SCIP team,
>>>>>>>
>>>>>>> I implemented a Constraint Handler for a special
>>>>>>> project scheduling problem.
>>>>>>>
>>>>>>> In the CONS_RESPROP function, I try to explain the bound
>>>>>>> changes processed through the propagation algorithm implemented in
>>>>>>> the CONS_PROP function.
>>>>>>>
>>>>>>> In the function for the explanation generation, I observed a problem
>>>>>>> with the function SCIPvarGetLbAtIndex(...):
>>>>>>>
>>>>>>> In the propagation algorithm, the lower bound is propagated with
>>>>>>> the function SCIPinferVarLbCons(). I integrate the integer lower bound
>>>>>>> of the variable responsible for the processed bound change as an
>>>>>>> inferinfo in this function.
>>>>>>>
>>>>>>> In the CONS_RESPROP function, I compare the latter value with the actual
>>>>>>> lower bound of the responsible variable given by
>>>>>>> SCIPvarGetLbAtIndex(...) at the given bound change index.
>>>>>>>
>>>>>>> My problem is that these values are not the same, which should not be
>>>>>>> the case in my understanding.
>>>>>>>
>>>>>>> Thank you in advance for your help.
>>>>>>>
>>>>>>> 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
>>>>>
>>>>
>>>
>>>
>>
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip
>

-- 
____________________________________________________________
Ambros M. Gleixner
Zuse Institute Berlin - Matheon - Berlin Mathematical School
http://www.zib.de/gleixner