How can I select the candidate interested in the solution pool?

Answered

Xianbang Chen

March 15, 2025 01:13

• 
• 
• 

Hi Gurobi guys,

I am using Gurobi + JuMP to handle a MILP model.

Below is my setting:

set_optimizer_attribute(jump_model, "PoolSearchMode", 2)

set_optimizer_attribute(jump_model, "PoolGap", 0.01)

After solving the MILP model, I obtained a pool of qualified solutions (let's say I have 10 qualified solutions), and I can retrieve them as needed.

My question is whether Gurobi provides any methods to select the solution I'm interested in, for example, the 5th solution, as the active solution, so that when I use codes like "value.(solution)," it always returns the 5th solution as desired.

Whether set_optimizer_attribute(jump_model, "SolutionNumber", 5) may work?

Thank you for any hints!

Best,

Xianbang

 

Related to

• Solution Pools
• Julia
• Attribute

0

• 

  Xianbang Chen

  March 15, 2025 16:57 Edited

  • 
  • 
  • 

  Below is a demo code. How to make ``value.(x)`` always output ``sol3``?

  using JuMP, Gurobi, MathOptInterface
  
  const MOI = MathOptInterface
  
  n = 5;
  
  capacity = 10.0;
  
  profit = [5.0, 3.0, 3.0, 7.0, 4.0];
  
  weight = [2.0, 8.0, 8.0, 2.0, 5.0];
  
  jump_model = Model(Gurobi.Optimizer)
  
  set_optimizer_attribute(jump_model, "PoolSearchMode", 2)
  
  set_optimizer_attribute(jump_model, "PoolGap", 0.4)
  
  @variable(jump_model, x[1:n], Bin)
  
  @constraint(jump_model, sum(weight[i] * x[i] for i in 1:n) <= capacity)
  
  @objective(jump_model, Max, sum(profit[i] * x[i] for i in 1:n))
  
  print(jump_model)
  
  optimize!(jump_model)
  
  solution_summary(jump_model)
  
  all_vars = JuMP.all_variables(jump_model)
  
  num_solutions = result_count(jump_model)
  
  on_vars = filter(v -> startswith(JuMP.name(v), "x"), all_vars)
  
  sol_1 = Dict(JuMP.name(v) => JuMP.value(v; result = 1) for v in on_vars)
  
  sol_2 = Dict(JuMP.name(v) => JuMP.value(v; result = 2) for v in on_vars)
  
  sol_3 = Dict(JuMP.name(v) => JuMP.value(v; result = 3) for v in on_vars)
  
  sol_4 = Dict(JuMP.name(v) => JuMP.value(v; result = 4) for v in on_vars)
  
  sol_5 = Dict(JuMP.name(v) => JuMP.value(v; result = 5) for v in on_vars)
  
  value.(x)

  0
• 

  Riley Clement

  • Gurobi Staff

  March 16, 2025 23:54

  Hi Xianbang,

  The attributes Xn and SolutionNumber should help here.

  Please see the discussion on Solution Pools for more context.

  If there's any difficulties accessing these through JuMP then please reach out to their discourse channel: https://discourse.julialang.org/tag/jump

  - Riley

  0
• 

  Xianbang Chen

  March 17, 2025 00:03

  • 
  • 
  • 

  Hi Riley,

  Thank you for your help. Yes, I read the files and tried a few things. Based on my experience, SolutionNumber doesn’t seem to work in JuMP. If I set SolutionNumber before optimize!, the model only stores the last solution, so SolutionNumber has no effect. If I set it after optimize!, it throws an error.

  I also tried:

  MOI.set(backend_model, MOI.VariablePrimal(), JuMP.index(v), val_3)
  

  to see if I could modify the optimal solution directly in the model, but that failed too.

  My final approach was to add equality constraints to force the binary variables to match the solution I wanted, then run optimize! again.

  Anyway, I appreciate your help.

  Best,
  Xianbang

  0
• 

  Riley Clement

  • Gurobi Staff

  March 17, 2025 00:09

  Hi Xianbang,

  If you were using a Gurobi API you would set SolutionNumber after optimization, then use the Xn attribute on variables (or the model) to retrieve the value.  I'm not sure setting SolutionNumber before optimization would have an effect.

  In any case JuMP seems to be doing it's own thing.  Maybe the following page will help? https://jump.dev/JuMP.jl/stable/manual/solutions/#Multiple-solutions

  - Riley

  0
• 

  Xianbang Chen

  March 17, 2025 00:18

  • 
  • 
  • 

  Hi Riley,

  I’m using a Julia package where JuMP handles the modeling. It seems JuMP doesn’t support SolutionNumber, but it’s helpful to know that Gurobi does—I’ll try using that feature when I work directly with the Gurobi API.

  For now, I followed the example you shared and wrote something like:

  for i in 1:result_count(model)
      println("Solution $i")
      println(" x = ", value.([x1, x2]; result = i))
  end
  

  This allows me to pull all solutions from the solution pool. Then, I can add constraints like binary_1 = 1 to force the model to use the solution I want as the active one.

  Thanks again for your help.

  Best,
  Xianbang

  0

Please sign in to leave a comment.