[Scip] Dual values of constraints of own constraint handler

[Grit Claßen]

Hello Stefan,

thank you for your detailed answer! Obviously, I had missed some details 
on the implementation of a branching constraint handler.

I have just one further question. Why is it important to set the 
"modifiable" flag of the rows to FALSE? Assume that I have created two 
child nodes of the root node, each containing a branching constraint 
which includes priced variables. The B&B procedure processes one of 
these two child nodes first. During this process new variables are 
priced which also have to be considered in the branching constraint of 
the second child node. This is why I would set this branching constraint 
to modifiable and add newly priced variables to the constraint before 
processing the node. I cannot do this when the constraint is not 
modifiable, can I?

Thank you for your assistance,
Grit

On 06/23/2012 12:27 PM, Stefan Heinz wrote:
> Hi Grit,
>
> the branching constraints should be initial. That means that the 
> scip_initlp() method of the C++ constraint handler class is called for 
> the node where the branching constraint is added. The initial flag is 
> the one in SCIPcreateCons()
> http://scip.zib.de/doc/html/scip_8h.html#a8d771b778ea2599827e84d01c0aceaf2 
>
> which you should call in you constraint handler. That ensurs that you 
> get a chance to put your row into the LP.
>
> In that callback (scip_initlp()) you create the SCIP_ROW* of your 
> constraint. It is important that the "local" flag is set to TRUE and 
> the "modifiable" and "removable" flag is set to FALSE. Setting the 
> removable flag to FALSE ensures that SCIP does remove the constraint 
> from the LP (except it leave the the sub tree). See the following two 
> methods for creating a row there you also see the three flags.
>
> SCIPcreateRow()
> http://scip.zib.de/doc/html/scip_8h.html#a0f3f5d2982142da67c59fb8de0ccbd9c 
>
>
> SCIPcreateEmptyRow()
> http://scip.zib.de/doc/html/scip_8h.html#ad250ffa5f36870fc0eb2c10e1e2eaa67 
>
>
> After that you add the row to the LP. Here it is important to set the 
> force flag to TRUE to ensure that the row enters the LP. Therefore use 
> the method:
> SCIPaddCut()
> http://scip.zib.de/doc/html/scip_8h.html#a04afcc23c6c21a921cd5e9e2f86d77ea 
>
>
> Doing all that should ensure that the branching constraint belonging 
> to the path of a particular search node are part of the LP relaxation. 
> The next step would be to get these active branching constraints. That 
> are the once of the path to a search node. That can be done in the 
> same fashion as in the binpacking example. You just ask for all active 
> constraints of your branching constraint handler. Therefore check the 
> code of the binpacking example:
>
> http://scip.zib.de/doc/examples/Binpacking/pricer__binpacking_8c_source.html#l00103 
>
>
> Having these active constraints, the final step is get the dual 
> values. Therefore, you should implement an interface method in you 
> constraint handler similar to once for the linear, knapsack, setppc, 
> and logicor constraint handler. Therefore have a look at the following 
> code.
> http://scip.zib.de/doc/html_devel/cons__logicor_8c_source.html#l03489
>
> This interface method delivers the dual values of the active constraints.
>
> Hope that helps
> Stefan
>
>
>
>
>
> On 06/22/2012 11:19 AM, Grit Claßen wrote:
>> Hello,
>>
>> this email somehow continues my question on SCIP_ConsData from the
>> beginnig of this week. I have implemented the constraint handler for
>> branching constraints in my Branch-and-Price framework. But now in my
>> pricer, I also have to consider the new constructed branching
>> constraints with their dual values. The question is how to get the dual
>> values of the constraints of my branching constraint handler?
>>
>> I have included a "SCIP_ROW* row" member in my SCIP_ConsData to be able
>> to call SCIProwGetDualsol(...). However, I do not know when the rows are
>> created. Do I have to create them in a callback function of the
>> constraint handler? I tried to create the rows when creating the
>> branching constraints. But there must be an error somewhere since my
>> program creates the same branching constraints twice (consecutively). I
>> presume that the rows are not handled correctly.
>>
>> An alternative way I can think of is to create the rows in the sepalp
>> callback. Would this be the correct way? Anyway, do I have to handle the
>> rows to be able to get dual values of my branching constraints?
>>
>> Thank you and best wishes
>> Grit
>>
>> -- 
>> Dipl.-Comp.Math. Grit Claßen
>>
>> UMIC Research Centre and Lehrstuhl II für Mathematik
>> RWTH Aachen University
>> 52056 Aachen
>> Germany
>>
>> phone: +49 241 80 20753 or +49 241 80 94999
>> http://www.math2.rwth-aachen.de
>> http://www.isek.rwth-aachen.de
>>
>> _______________________________________________
>> Scip mailing list
>> Scip at zib.de
>> http://listserv.zib.de/mailman/listinfo/scip
>
>


-- 

Dipl.-Comp.Math. Grit Claßen

UMIC Research Centre and Lehrstuhl II für Mathematik
RWTH Aachen University
52056 Aachen
Germany

phone: +49 241 80 20753 or +49 241 80 94999
http://www.math2.rwth-aachen.de
http://www.isek.rwth-aachen.de