Wrapper for Gurobi var?

Answered

Bruno Colonetti

February 21, 2023 01:53

• 
• 
• 

Hello, everyone

I'm writing a Benders algorithm and there is a programming problem that I just cannot solve.

As you know, Benders is an iterative method that goes back and forth between a master problem and a subproblem. Its upper bound (for minimization problems) is not necessarily decreasing so you could very well find one optimal solution in the first iteration and spend the rest of the time just improving the LB. 

The thing is: is there any way to write a wrapper for Gurobi vars in Python with a x property that instead of returning the vars' x of the last time the model was run, it returns the vars' x for the best solution found in BD? What I'm trying to do is to create a wrapper that behaves just as a var except when the x property is accessed, so I can seamlessly switch between BD and a direct approach with Gurobi without having to change too much of the code. I would say that this would be generally possible in Python, but I'm not sure it is possible with gurobipy. Whenever I try to inherent gurobipy.Var, I get a RecursionError.

P.S: reoptimizing the models is not an option. I'm also open to suggestions other than wrappers.

Thanks in advance!
Bruno

Related to

• Python
• Bender's

0

• 

  Riley Clement

  • Gurobi Staff

  February 22, 2023 00:24

  Hi Bruno,
  
  An alternative to extension through inheritance, is composition.  You can also make use of automatic delegation.

  However trying to use either approach, to substitute an object from a 3rd party API (such as gurobipy) is very risky as there may be type checking under the hood, and cause unexpected behavior that you do not realize.

  Would it be possible in your case to just define a method to switch between the solution values and monkey patch it on (for aesthetics)?  E.g.

  def set_BD_solution(var, x):
      var._BD_solution = x
  
  def get_BD_solution(var):
      return var._BD_solution
  
  def switch_X(var, BD=False):
      if BD:
          return get_BD_solution(var)
      return var.X
  
  # monkey patch if required
  gp.Var._X = switch_X
  gp.Var._set_BD_solution = set_BD_solution
  gp.Var._get_BD_solution = get_BD_solution
  
  # define variables after monkeypatching
  
  # for a variable "var" access like so
  var._X()  # optimal solution
  var._X(BD=True)  # Benders solution

  
  - Riley

  1
• 

  Bruno Colonetti

  June 12, 2023 00:06

  • 
  • 
  • 

  Hey, Riley
  
  Thanks for the tip! At the end, I chose to follow your advice and it's working just fine
  
  Thanks!

  0

Please sign in to leave a comment.