Can model.copy() also copy my own data structures of variables in the original model (in Python)?

Answered

Bo Lan

• September 20, 2019 17:42

Greetings,

I made a model and added binary decision variables  into a data structure like the following in Python:

def make_model(some_arguments):

      model = grb.Model()

      ......

      x[i,j] = model.addVar(......)

      y[i,j] = model.addVar(......)

      ......

      model._data = x, y

      return model

 

Then in another function I can extract those groups of variables, i.e. x, y, like

def solve_model(model):

      x, y = model._data

      ......

 

Now I want to make a copy of the original model and manipulate the copy and solve it again. the code is something like

copy = model.copy()

This is a good copy of the original  model but it does not copy my data structure of x, y. When I call the following function

def solve_copy(copy):

      x, y = copy._data

     for i in y:
          y[i].ub = 0

      ......

The line of 'x, y = copy._data' will throw an error of 'AttributeError: 'gurobipy.Model' object has no attribute '_data''. then I cannot manipulate y with their upper bounds.

 

Can anyone give me some suggestions about how to copy a model and its data structures of decision variables  simultaneously?

 

Thanks,

 

Larry

 

0

• 

  Baptiste Lambin

  • January 10, 2020 14:42

  

  

  Hello there,

  I don't know the policy of this forum regarding duplicate posts.

  To avoid posting the same topic, I thus post here to see if anyone has an answer to this, as I also would like to be able to copy a model together with its custom attributes (attributes starting with _)

  Thanks a lot,
  Regards,
  Baptiste

  0
• 

  Matthias Miltenberger

  Gurobi Staff

  • January 15, 2020 16:27

  Hi Larry and Baptiste,

  I don't suppose there is a way currently to copy user data when calling Model.copy(). I recommend to only use one data object _data to store all your user data and after copying the model call model2._data = model._data.

  Cheers,
  Matthias

  0
• 

  Baptiste Lambin

  • January 15, 2020 16:45

  

  

  Hi Matthias,

  Thanks for your answer.

  However for the case described here, using model2._data = model._data will not work, as model._data contains variables from model, model2._data would also contain the variables from model. Here is a minimal working example in Python :

  m = Model()
  x = m.addVar(vtype=GRB.BINARY, name="x")
  y = m.addVar(vtype=GRB.BINARY, name="y")
  m.update()
  m._data = [x,y]
  m2 = m.copy()
  m2._data = m._data
  x2,y2 = m2._data
  m2.addConstr(x2 + y2 == 1)

  The addConstr call then throw an error "GurobiError: Variable not in model".
  I am not surprised by this though, but it would be nice to be able to make it work "easily".
  Of course, you can always create the correct m2._data using
  m2._data = [m2.getVarByName("x"), m2.getVarByName("y")]
  but it is a bit cumbersome.

  The usecase I was thinking about is e.g. callbacks, where using custom attributes makes it really easy to access the variables inside the callback. Having the correct variables directly in m2._data make it easier to use e.g. two different callbacks.

  But I'll use getVarByName to get it done, the current behavior makes sense in a way, I was just wondering if there was some obscure way to easily do this.

  Thanks again,
  Regards,
  Baptiste

   

  0
• 

  Bo Lan

  • January 15, 2020 20:56

  

  

  

  Thank you, Baptiste, for sharing your practice. It's helpful.

  Larry

  0

Please sign in to leave a comment.