Time needed to create gurobi constraints is too long

Answered

Panggah Prabawa

• July 31, 2023 04:33

Hi,

I recently stumbled on the creation time of constraints using model.addConstrs().

p_line = {(i,j,t): m.addVar(vtype=GRB.CONTINUOUS, lb=-100, ub=100, name='P_line({}-{}),t={}'.format(i,j,t)) for t in set_t for i in (inner_bus | substation) for j in set_tbus[i]}
q_line = {(i,j,t): m.addVar(vtype=GRB.CONTINUOUS, lb=-100, ub=100, name='Q_line({}-{}),t={}'.format(i,j,t)) for t in set_t for i in (inner_bus | substation) for j in set_tbus[i]}
l_line = {(i,j,t): m.addVar(vtype=GRB.CONTINUOUS, lb=0, ub=100, name='L_line({}-{}),t={}'.format(i,j,t)) for t in set_t for i in (inner_bus | substation) for j in set_tbus[i]}

v_sq = {(i,t): m.addVar(vtype=GRB.CONTINUOUS, lb=0.9025, ub=1.1025, name='v_square({}),t={}'.format(i,t)) for t in set_t for i in set_bus}

VoltageDifference = m.addConstrs((v_sq[i,t] - v_sq[j,t] == 2*(R(net,i,j)*p_line[i,j,t]+X(net,i,j)*q_line[i,j,t]) - (R(net,i,j)**2+X(net,i,j)**2)*l_line[i,j,t] for i in (inner_bus | substation) for j in set_tbus[i] for t in set_t),name='Voltage_difference')

# R and X are parameters,
# p_line, q_line, and l_line are tuples (32,32,96)
# v_sq is a tuple (33,96)

I formulated the model using Jupyter Notebook inside VSCode with Python 3.7. It takes 14 seconds to create this set of constraints using the model.addConstrs() or tupledict methods. I know the quickest way is to use the matrix form to model the formulation. However, there are many other high-dimensional constraints that I already constructed (they have low creation time) which will need a lot of time to reconstruct in matrix form, especially when there are nonlinear ones.

 

Is it related to the variables which are in the tuple form?

Do my Python and Gurobi versions matter?

Is there any workaround for this problem?

 

I provided the screenshot of my problem below.

 

Thank you!

Regards, Panggah.

Related to

• Performance

1

• 

  Jaromił Najman

  Gurobi Staff

  • August 01, 2023 09:01

  Hi Panggah,

  Is it related to the variables which are in the tuple form?

  You construct your variables as dictionaries so this should be fine.

  Do my Python and Gurobi versions matter?

  They may matter. It is best to always use the latest Gurobi version if possible. To see which Python version is supported, please refer to Support Platforms.

  Is there any workaround for this problem?

  Very likely there is one. However, it is hard to tell from the small code snippet. Could you please provide a minimal reproducible example?

  Best regards, 
  Jaromił

  0
• 

  David Torres Sanchez

  Gurobi Staff

  • August 01, 2023 09:59

  Hi Panggah,

  Do my Python and Gurobi versions matter?

  I would always try and use the latest versions for both.

  Is there any workaround for this problem?

  Have you tried using \(\texttt{model.addConstr}\) (this can sometimes be faster):

  VoltageDifference = {
      (i, j, t): m.addConstr(
          (
              v_sq[i, t] - v_sq[j, t]
              == 2 * (R(net, i, j) * p_line[i, j, t] + X(net, i, j) * q_line[i, j, t])
              - (R(net, i, j) ** 2 + X(net, i, j) ** 2) * l_line[i, j, t]
          ),
          name=f"Voltage_difference[{i},{j},{t}]",
      )
      for i in (inner_bus | substation)
      for j in set_tbus[i]
      for t in set_t
  }

  Cheers, 
  David

  0
• 

  Panggah Prabawa

  • August 01, 2023 10:10

  

  

  

  Hi, Jaromił.
  Thank you for your reply.

  They may matter. It is best to always use the latest Gurobi version if possible. To see which Python version is supported, please refer to Support Platforms.

  I am using the latest Gurobi version, but my Python is still v3.7 because I have some other libraries that force me to use it for my project. I think the Python version will not affect much, but I will give it a try.

  Very likely there is one. However, it is hard to tell from the small code snippet. Could you please provide a minimal reproducible example?

  I will provide you the code later for you to try :)

  Thank you for your suggestions.

  Regards,
  Panggah

  0
• 

  Panggah Prabawa

  • August 01, 2023 10:24
  • Edited

  

  

  

  Hi, David.
  Thank you for your reply.

  VoltageDifference = {
      (i, j, t): m.addConstr(
          (
              v_sq[i, t] - v_sq[j, t]
              == 2 * (R(net, i, j) * p_line[i, j, t] + X(net, i, j) * q_line[i, j, t])
              - (R(net, i, j) ** 2 + X(net, i, j) ** 2) * l_line[i, j, t]
          ),
          name=f"Voltage_difference[{i},{j},{t}]",
      )
      for i in (inner_bus | substation)
      for j in set_tbus[i]
      for t in set_t
  }

  I exactly tried this one before I posted my problem. The creation time did not differ much, though. Below is the screenshot :)

  

  It is quite a drawback if creating all the constraints takes almost 20 seconds while it takes only 0.3 seconds to solve. I need a quick running time that covers both creating and solving the model for the purpose of putting it into an iterative algorithm (around 1000+ iterations). As Jaromił said, I will also provide a reproducible example.

  Thank you for your kind suggestion!

  Regards,
  Panggah

  0
• 

  Panggah Prabawa

  • August 02, 2023 01:38

  

  

  

  I think I found the culprit which is not related to Gurobi at all.

  The R(net,i,j) and X(net,i,j) are basically very simple functions, yet takes a little while to call.

  

  The R and X are set to a constant in the above picture, and the problem appears no more.

  I think I do not need to provide a reproducible code as it was my own wrongdoing and has nothing to do with Gurobi :)

  I am sorry for my mistake and thank you for your kind suggestions.

  Regards,
  Panggah

  0
• 

  Jaromił Najman

  Gurobi Staff

  • August 02, 2023 07:43

  I am sorry for my mistake and thank you for your kind suggestions.

  No problem at all and it's great to hear that you found the source of this issue.

  0

Please sign in to leave a comment.