Finding all optimal solution for a Multi-objective Optimization

Answered

Purnima Subramaniam

• March 11, 2023 07:42

Hi,

I have set up a multi-objective mixed integer optimization model with Gurobi. In my case, I have two objectives. As far as I have understood, Gurobi does not support finding a Pareto frontier. I have two questions:

1.) I was trying to find multiple optimal solution using the PoolSearchMode and PoolSolutions. I thought this way gurobi will give me all possible optimal solutions in the feasible region. And I would use these point to find a Pareto frontier. Is it a right approach ?

2.) I am unable to retrieve multiple optimal solutions and get the program to print all the optimal values for both objective functions. This is the code I am using to print all the values, but it is throwing error.

m.Params.PoolSearchMode = 2
m.Params.PoolSolutions = 5
m.Params.PoolGap = 1
         
m.optimize()
m.solCount

for k in range(m.SolCount):
    m.Params.SolutionNumber = k
    print(Obj1.getValue())
    print(Obj2.getValue())
   

This is not printing all feasible solutions for for both objective value. Please help.

 

Thanks,

Purnima

 

Related to

• Multiple Objectives
• Solution Pools

0

• 

  Marika Karbstein

  Gurobi Staff

  • March 13, 2023 12:40

  Hi Purnima,

  You need to set the parameter for each objective. If you have 2 objectives, you can access the environment for both objectives and then set parameters as follows

  env0 = m.getMultiobjEnv(0)
  env1 = m.getMultiobjEnv(1)
  # setting pool parameter for first objective
  env0.setParam('PoolSearchMode', 2)
  env0.setParam('PoolSolutions', 10)
  env0.setParam('PoolGap', 1)
  # setting pool parameter for second objective
  env1.setParam('PoolSearchMode', 2)
  env1.setParam('PoolSolutions', 10)
  env1.setParam('PoolGap', 1)
  # allow the complete pool to contain 10 solutions from first objective
  # and 10 solutions from second objective
  m.setParam('PoolSolutions', 20)

  In this way, Gurobi will search for 10 solutions when optimizing the first objective and 10 solutions when optimizing the second objective. Note that the model with the second objective also contains the constraint for the allowed degradation of the first objective.

  When checking the objectives for all solutions, you probably need to set the parameter ObjNumber in a similar way as you set SolutionNumber.

  I hope this helps,
  Marika

  0
• 

  Purnima Subramaniam

  • March 13, 2023 17:35

  

  

  

  Hi Marika,

  Thank you for your response. It's still not clear to me on how to use ObjNumber to print all the values of optimal solution for both Obj1 and Obj2.
  
  

  m.optimize()
  m.solCount

  for w in range(m.SolCount):
      m.Params.SolutionNumber = w
      print([var.X for var in Obj1.getValue()])
      print([var.X for var in Obj2.getValue()])
      

  If I try to run this piece of code , I get this error : AttributeError: 'float' object has no attribute 'X'

  Could help me figure out how to print all the solution values ?

   

  Thanks,

  Purnima

  0
• 

  Marika Karbstein

  Gurobi Staff

  • March 13, 2023 18:45
  • Edited

  Hi Purnima,

  You can print the objective values as follows

  for w in range(m.SolCount):
      m.Params.SolutionNumber = w
      # consider all objectives for this solution
      for i in range(2):
          m.Params.ObjNumber = i
          # value of objective i
          print(m.ObjNVal)
          
      # print all variable values for this solution
      m.printAttr('Xn')

  Best regards,
  Marika

  0

Please sign in to leave a comment.