Copy a SCIP Model

[adecholaA1]

Hello,
I'm encountering an error when trying to copy the scip model (following the directions in the documentation). My aim is to copy the current state of the scip model, then add a constraint and obtain information such as the value of the LP objective function and the status of the model at the end of this resolution independently (without modifying the state of the current scip model). Proceeding as described in the code below, I get the message “zsh: segmentation fault”. Is there an error in the way I'm proceeding? Is there another way of copying the state of a scip model to which to make changes, resolve it and obtain information without modifying the original scip model?

Thank you in advance.

from pyscipopt import Model

class Copy_scip():
    def __init__(self, scip):
        self.scip = scip

    def getInfosUsingCopyScip(self, list_1, rhs):

        # Get back to the original problem 
        self.scip.freeTransform()

        # Model copy
        scip_copy = Model(sourceModel=self.scip)

        # Copy model variables
        cols = scip_copy.getLPColsData()
        scip_vars = [col.getVar() for col in cols]

        # Add constraint
        expr = sum(c*v for c, v in zip(list_1, scip_vars))
        scip_copy.addCons(expr <= rhs)

        #Optimize
        scip_copy.hideOutput()
        scip_copy.optimize()

        status = scip_copy.getStatus()
        lperror = (status=="lperror")
        cutoff = (status == "infeseable")
        LPObj = scip_copy.getLPObjVal() 

        scip_copy.freeProb()

        return LPObj, cutoff, lperror

if __name__=="__main__":
    scip = Model()
    x1 = scip.addVar("x1", vtype="I", lb=0)
    x2 = scip.addVar("x2", vtype="I", lb=0)

    scip.addCons(-x1 - 2*x2 <= -1)
    scip.addCons(-x1 + 2*x2 <= 1)
    scip.addCons(x1  + 2 * x2 <= 3)
    scip.addCons(x1  - 2 * x2 <= 1)

    scip.setObjective(2* x1 - x2, sense="minimize")

    #----------------------------------------

    copy = Copy_scip(scip)
    rhs = 1
    list_1 = [1.2, 2.5]

    LPObj, cutoff, lperror = copy.getInfosUsingCopyScip(list_1, rhs)

[Joao-Dionisio]

Hello, @adecholaA1! The method getLPColsData is calling the SCIPgetLPColsData function, which can only be called during the SOLVING stage. By the time you're trying to call it, you're in the PROBLEM stage. There should be an error raised, though, I'm unsure why it's not. It should be the same problem causing #971.

Instead of getLPColsData, you can simply use scip_copy.getVars. If you prefer to work with LPs directly, you can take a look at the LP interface, although it's not as ergonomic, and the possibilities are limited.

[adecholaA1]

Thanks for your answer. I replaced getLPColsData, as you suggested, and it works well—no error. However, I still get the segmentation fault message when trying to use optimize on the copied model. I think it is the same internal issue within SCIP. I'm not sure if it's possible to optimize the copied model while the original model is still in the SOLVING stage.

[Joao-Dionisio]

Can you give a small example of the problem you're having now? From what I see in the documentation, you should be able to call optimize if a model is in the SOLVING stage, but maybe it being a copied model messes things up somehow.

[adecholaA1]

I want to optimize the copied model and retrieve some information, such as the status and the value of the LP objective function. However, as in the code above — specifically in the following part:

# Optimize
scip_copy.hideOutput()
scip_copy.optimize()

I get a segmentation fault.

[Joao-Dionisio]

After replacing the getLPColsData with scip_copy.getVars() in the code above, I no longer get any issues. Can you provide the code that's not working as intended?

[Joao-Dionisio]

Any updates @adecholaA1?

[adecholaA1]

Hello @Joao-Dionisio ,
I apologize for my late reply. I finally gave up on this avenue. Thank you for your help with this issue.