Slow model building

Awaiting user input

Vegard Hansen

August 13, 2024 00:11

• 
• 
• 

Hello,

I am building a branch-and-price application, and my master-problem is taking a very long time to build (> 10 minutes) (and I am getting RAM memory issues after a while when building it), however actually solving the problem is super fast (often < 1 s)

It has the following data rows, columns and entries:
Optimize a model with 598 rows, 5000 columns and 691142 nonzeros
Model fingerprint: 0x040fac46
Coefficient statistics:
  Matrix range     [1e+00, 1e+09]
  Objective range  [2e+05, 1e+08]
  Bounds range     [1e+00, 1e+00]
  RHS range        [1e+00, 2e+09]
Warning: Model contains large matrix coefficients
Warning: Model contains large rhs
         Consider reformulating model or setting NumericFocus parameter
         to avoid numerical issues.

The formulation is as follows:

problem.setObjective(quicksum((revenue_column[(g,k)] + end_of_horizon_value_column[(g,k)] - harvest_cost_column[(g,k)]) * l[(g,k)] for g in all_generations for k in production_schedules[g]), GRB.MAXIMIZE)

    

    for t in range(n_timeperiods):

        problem.addConstr(quicksum(total_biomass_column[(g,t,k)] * l[(g,k)] for g in relevant_generations[t] for k in production_schedules[g]) <= MAB,

                            "MAB_limit[%s]" %(t))

       

    for g in all_generations:

        if g != all_generations[-1] and g != all_generations[-2]:

            problem.addConstr(quicksum(l[(g,k)] for k in production_schedules[g]) + quicksum(l[(g+1,k)] for k in production_schedules[g+1]) + quicksum(l[(g+2,k)] for k in production_schedules[g+2]) <= 1, "consecutive_generations[%s]" %(g))

    for t in range(n_timeperiods):

        for u in units:

            problem.addConstr(quicksum(inhabit_or_wash_column[(g,t,u,k)] * l[(g,k)] for g in relevant_generations[t] for k in production_schedules[g]) <= 1,

                            "only_one_gen_per_unit[%s,%s]" %(t,u))

 

    for g in all_generations:

        problem.addConstr(quicksum(l[(g,k)] for k in production_schedules[g]) == 1,

                            "convexity[%s]" %(g))

I realize I am dealing with a large, real-world issue, so indices are generations -> 100, timeperiods -> 100, units -> 20

However it still does not make fully sense to me that the model should actually be this slow... Any ideas would be greatly appreciated!

Related to

• Model Building

0

• Official comment

  

  Simranjit Kaur

  • Gurobi Staff

  November 13, 2026 01:21

  This post is more than two years old. Some information may not be up to date. For current information, please check the Gurobi Documentation or Knowledge Base. If you need more help, please create a new post in the community forum, or try Gurobot, our chatbot interface offering instant, expert-level support.
• 

  Riley Clement

  • Gurobi Staff

  August 13, 2024 23:00

  Hi Vegard,

  That does sound unusually slow.  Can you show us how you are creating the variables?

  - Riley

  0
• 

  Vegard Hansen

  August 13, 2024 23:26

  • 
  • 
  • 

  I am building the variables like this:

  l = dict.fromkeys((g,k) for g in all_generations for k in production_schedules[g])
      for g in all_generations:
          for k in production_schedules[g]:
              l[(g,k)] = problem.addVar(name = 'lambda({},{})'.format(g,k))
  
  The model building scales very badly with the nr of columns k, already at k > 50 it starts getting quite slow.

  0
• 

  Riley Clement

  • Gurobi Staff

  August 13, 2024 23:36

  Without seeing the entire code, and the data structures you are using it is hard to say what the cause is.  Are you using pandas anywhere?

  In your code snippet for building variables you can replace the first line with

  l = dict()

  and the last line with

  l[g,k] = problem.addVar(name = 'lambda({},{})'.format(g,k))

  The first thing I would do is invoke the golden rule of computers and turn it off and on again.  I'm wondering if you are running low on memory and a reset could help.

  The second thing I would do is record the time taken for each chunk of code, so you can narrow down where the issue is:

  import time
  
  # stuff here
  
  start_time = time.time()
  # build variables
  end_time = time.time()
  execution_time = end_time - start_time
  print(f"Variable creation time: {execution_time} seconds")
  
  start_time = time.time()
  # first set of constraints
  end_time = time.time()
  execution_time = end_time - start_time
  print(f"First constraints creation time: {execution_time} seconds")
  
  # and so on...

  - Riley

   

  0

Post is closed for comments.