[SCIP] Using SCIPdisableCons in problem creation stage

Fri Oct 25 13:12:45 CEST 2019

Hi Gregor,

Thank you for your insight. I found the problem and it is a mistake from me when setting up the variables.

Best,

Lan

________________________________
From: Gregor Hendel <hendel at zib.de>
Sent: 16 October 2019 15:20:01
To: Ngô Lan
Cc: Scip at zib.de
Subject: Re: [SCIP] Using SCIPdisableCons in problem creation stage

Hi Lan,

sorry for taking so long to respond. There are two concepts that I would like to mention. The first is the so-called effective root depth. During search, it may eventually happen that say, the left subtree of the root node has been completely searched. At this moment, the right child node effectively becomes the new root node, and therefore, the effective root depth of SCIP is increased by 1.

All constraints added to nodes at the effective root depth or even higher are considered global. You can check for the effective root depth through a call to SCIPgetEffectiveRootDepth() from scip_tree.h.

The second concept is the concept of the "validnode" argument of SCIPaddConsNode(). If the node where this constraint is valid is higher than the effective root depth, the constraint is marked as a global constraint.

Either way, this change happens through the method SCIPconsSetLocal(), which I suggest you watch within GDB (after compiling both your code and SCIP in debug mode) to trace back the source of your trouble.

Let me know what you find out,
Gregor

Am 11.10.19 um 19:03 schrieb Ngô Lan:

Hi Gregor,

For my problem, in each child node, I would add 2 constraints (1 linear and 1 from my own constraint handler as a marker constraint). When printing out in SCIP_DECL_CONSACTIVE, the right marker constraint is activated. For the linear constraint, I did create it as local constraint, save it in a vector in probdata and check it in my pricer but here when I used SCIPconsIsLocal, it returned False. I used also SCIPconsIsLocal directly after I created the linear constraint but here I got True. Is there something SCIP does internally that affects the local attribute?

Best,

Lan

________________________________
From: Gregor Hendel <hendel at zib.de><mailto:hendel at zib.de>
Sent: 11 October 2019 15:26:51
To: Ngô Lan
Cc: scip at zib.de<mailto:scip at zib.de>
Subject: Re: [SCIP] Using SCIPdisableCons in problem creation stage

Hi Lan,

but in the Binpacking example, everything works as expected. Both constraints added by ryanfoster-branching are inactive at first, because they are added to the child nodes. They are only activated when their nodes become active.

  1.  Did you accidentally create your constraints with the "local" flag set to FALSE?
  2.  Can you trace back when the constraint gets activated? You can do this by printing a nice debug message from within the SCIP_DECL_CONSACTIVE-callback of your constraint handler.

Best,
Gregor

Am 10.10.19 um 12:21 schrieb Ngô Lan:

Hi Gregor,

Yeah the second question is about the main scip. So I saved all the branching constraint to probdata after adding to main scip and later in subscip, going through the vector of branching constraints, checking which one is active using SCIPconsIsActive and getting the dual of those that are active. The issue here is that when I branch to 2 child nodes and enter the first child node, it says the the constraint for the other child node is also active in this node. I just followed what you did in binpacking example: create cons, create child, and add cons to child node. I think it is pretty clear so that is why I don't really have any ideas what can go wrong here. Any help is appreciated.

Thank you for your time.

Best,

Lan

________________________________
From: Gregor Hendel <hendel at zib.de><mailto:hendel at zib.de>
Sent: 10 October 2019 09:47:38
To: Ngô Lan; Scip at zib.de<mailto:Scip at zib.de>
Subject: Re: [SCIP] Using SCIPdisableCons in problem creation stage

Hello Lan,

find the answers to your two questions inline.

Am 10.10.19 um 01:40 schrieb Ngô Lan:

Hi Gregor,

Thank you for your reply. It works now. Would it be possible to add the deleted constraints again to scip? Ideally, when a node is entered again, I would like to enable the constraint instead of creating a new one. My code does not run yet so I haven't had chance to test it.

As long as you do not release the constraint pointers (using SCIPreleaseCons()), SCIP will not physically delete the constraint, but only remove it from the problem, so that you can later add them again. An explanation of the capture/release mechanism for variables can be found here https://scip.zib.de/doc/html/OBJ.php, the explanation also applies to constraints.

I also have another problem with the child nodes. So for branching, I created a linear branching constraint and a marker constraint from my constraints handler and added them into a child node (sticking at node is true). However, in subscip when I check in which branch I am at using SCIPconsIsActive, I get true even for the constraints that were not created in this node. Do you have any idea where things might go wrong?

As far as I understand your approach, all constraints of the subscip should be active, because at every node (of the main solving process), you free the transformed problem of the subscip and then you add exactly those constraints that reflect your current branching decision. Are you using SCIPconsIsActive on the main solving process constraints, or on the subscip constraints? For the main SCIP, I expect that all branching constraints (there might be several ones along the path from the root) are active, and the rest is not active.

Kind regards,
Gregor

Best,

Lan

________________________________
From: Scip <scip-bounces at zib.de><mailto:scip-bounces at zib.de> on behalf of Gregor Hendel <hendel at zib.de><mailto:hendel at zib.de>
Sent: 09 October 2019 10:32:17
To: scip at zib.de<mailto:scip at zib.de>
Subject: Re: [SCIP] Using SCIPdisableCons in problem creation stage

Good morning Lan,

how about using SCIPdelCons() instead of SCIPdisableCons()?

Let me know if that works for you,
Gregor

Am 09.10.19 um 01:03 schrieb Ngô Lan:

Hi all,

I am solving a project scheduling problem using branch-price-cut and I am having an issue with SCIPdisableCons function. So after each pricing iteration, I will not release the subscip but only free the transformed problem using SCIPfreetransform and modify the objective function of the subscip in the next iteration. During branching, in CONSACTIVE, I will create an additional constraint for the subproblem depending on the branching decision and in CONSDEACTIVE, I will disable the related constraint using SCIPdisableCons. However, I got the error: cannot call method <SCIPdisableCons> in problem creation stage. Do you know another function that can work? Or what should I do differently?

Thank you for your time,

Best,

Lan

_______________________________________________
Scip mailing list
Scip at zib.de<mailto:Scip at zib.de>
https://listserv.zib.de/mailman/listinfo/scip

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20191025/9c38b2fd/attachment.html>