The benchmark is adapted from the JuMP paper. Eight optimization models with different sizes are selected as test cases, including four facility location models and four linear quadratic control problems. We conduct two rounds of benchmark using Gurobi and COPT as optimizer respectively. For each model, we measure the total time of modeling interface to generate model and pass it to the optimizer, and the time limit of optimizer is set to 0.0 seconds to avoid the influence of solution process.
All code to run the benchmarks is available at https://github.com/metab0t/PyOptInterface_benchmark.
:::{table} Time (second) to generate model and pass it to Gurobi optimizer. :widths: auto :align: center
| Model | Variables | C++ | PyOptInterface | JuMP | gurobipy | Pyomo |
|---|---|---|---|---|---|---|
| fac-25 | 67651 | 0.2 | 0.2 | 0.2 | 1.2 | 4.1 |
| fac-50 | 520301 | 0.8 | 1.2 | 1.8 | 9.7 | 32.7 |
| fac-75 | 1732951 | 2.7 | 4.1 | 6.6 | 32.5 | 119.3 |
| fac-100 | 4080601 | 6.3 | 10.0 | 17.8 | 79.1 | 286.3 |
| lqcp-500 | 251501 | 0.9 | 1.5 | 1.3 | 6.3 | 23.8 |
| lqcp-1000 | 1003001 | 3.7 | 6.0 | 6.1 | 26.7 | 106.6 |
| lqcp-1500 | 2254501 | 8.3 | 14.0 | 17.7 | 61.8 | 234.0 |
| lqcp-2000 | 4006001 | 14.5 | 24.9 | 38.3 | 106.9 | 444.1 |
:::
:::{table} Time (second) to generate model and pass it to COPT optimizer. :widths: auto :align: center
| Model | Variables | C++ | PyOptInterface | JuMP | coptpy | Pyomo |
|---|---|---|---|---|---|---|
| fac-25 | 67651 | 0.3 | 0.2 | 0.3 | 0.6 | 4.1 |
| fac-50 | 520301 | 2.2 | 1.5 | 2.7 | 5.4 | 32.8 |
| fac-75 | 1732951 | 8.1 | 6.6 | 10.2 | 20.3 | 117.4 |
| fac-100 | 4080601 | 22.4 | 23.4 | 30.3 | 58.0 | 284.0 |
| lqcp-500 | 251501 | 3.8 | 3.1 | 3.0 | 6.6 | 26.4 |
| lqcp-1000 | 1003001 | 16.0 | 15.5 | 13.9 | 28.1 | 112.1 |
| lqcp-1500 | 2254501 | 37.6 | 32.4 | 33.7 | 64.6 | 249.3 |
| lqcp-2000 | 4006001 | 68.2 | 60.3 | 66.2 | 118.4 | 502.4 |
:::
Recently, there are a lot of requests to test the performance of PyOptInterface compared with linopy and cvxpy, so we prepare a benchmark.
This is the result of benchmark, where the performance of PyOptInterface exceeds linopy and cvxpy significantly.
:::{table} Time (second) to generate and solve a linear programming model with Gurobi optimizer. :widths: auto :align: center
| N | Variables | PyOptInterface | linopy | cvxpy |
|---|---|---|---|---|
| 100 | 20000 | 0.112849 | 0.422408 | 0.373407 |
| 200 | 80000 | 0.294830 | 1.118702 | 1.575949 |
| 300 | 180000 | 0.710237 | 2.462809 | 4.038862 |
| 400 | 320000 | 1.256276 | 4.535225 | 8.687895 |
| 500 | 500000 | 2.189127 | 8.243707 | 18.941519 |
:::
We use the AC Optimal Power Flow problem to benchmark performance of PyOptInterface against different modeling languages. The code is released at GitHub and the result is published by our paper Accelerating Optimal Power Flow with Structure-aware Automatic Differentiation and Code Generation.