pyomo-MINLP-benchmarking

What is this?

This repository mostly contains the run_benchmarks.py script as well as a collection of MINLP problems/models, which have been taken from the MINLPlib and converted to Pyomo models with the translate.sh file.

Example usage

Translate gms model to pyomo model

1. Download the gms file from the MINLPlib.

2. Create a txt file(like convex.txt) to denote the instances that you want to translate.

3. Check the INSTANCEDIR, TESTSET and PYOMOMODELDIR in translate.sh and run it.

Benchmark

When running

# show some help
python run_benchmarks.py -h  

python run_benchmarks.py --solver mindtpy --model-dir models_all_solvable2 --timelimit 900 --mip-solver cplex --nlp-solver ipopt --iteration-limit 2000 --strategy OA --method-name direct_cplex_8threads --threads 8 --stalling-limit 2000 --result-folder LOA

# use subsolver from gams
python run_benchmarks.py --solver mindtpy --model-dir models_all_solvable2 --timelimit 900 --mip-solver cplex --nlp-solver gams --nlp-solver-args '{"solver":"ipopth"}' --iteration-limit 2000 --strategy OA --method-name direct_cplex_8threads --threads 8 --stalling-limit 2000 --result-folder LOA

# skip folder
python run_benchmarks.py --solver mindtpy --model-dir models_all_solvable2 --timelimit 900 --mip-solver cplex --nlp-solver gams --nlp-solver-args '{"solver":"ipopth"}' --iteration-limit 2000 --strategy OA --method-name no_regularization_direct_cplex_8threads --threads 8 --stalling-limit 2000 --result-folder LOA --skip-folder LOA/mindtpy-OA-2021_02_14_12_03_32 --no-skip-failed

After running, the results/<solver> dir will contain

• a .txt file for each model output
• trace_file.trc which can be loaded into Paver to generate automatic benchmarking plots (not yet tested)
• solving_times.csv which contains the model name, aswell as the solving time or the termination condition/error