BO4IO

This repository contains codes and data for the BO4IO paper (here). The following are the instructions for running the codes.

Citation

@misc{lu2024bo4io,
    title={BO4IO: A Bayesian optimization approach to inverse optimization with uncertainty quantification}, 
    author={Yen-An Lu and Wei-Shou Hu and Joel A. Paulson and Qi Zhang},
    year={2024},
    eprint={2405.17875},
    archivePrefix={arXiv},
    primaryClass={math.OC}}

Requirements

Dependencies can be found in the environment.yml file.

We use Gurobi as the primary optimization solver. here

Usage

To reproduce the results for each case study in the paper, clone the repository, install the required package and solver, and follow the instructions below.

Steps to run the FBA case study

This case study aims to learn the cellular objective in the FBA problem. All codes should be executed under the FBA folder.

1. Run BO4IO under specific problem settings

   python BO4IO_FBA.py -nexp 50 -nrxn 3 -ntrial 5 -niter 250 -ninit 5 -n_data_seed 1 -init_seed 1 -mp 1 -n_p 10 -sigma 0.1

   Description of program inputs:

   • nexp - number of training experiments (FOPs)
   • nrxn - number of reactions (common cellular objectives) included in the multiobjective, i.e. dimensions of unknown parameters
   • ntrial - number of BO trials
   • niter - number of BO iterations
   • ninit - number of initial samplings for building GP
   • n_data_seed - number of random instances to be run in the program
   • init_seed - the starting seed of the random instance
   • mp - with (1) or without (0) multiprocessing/parallelization
   • n_p - number of parallelized workers
   • sigma - noise level on observed decisions

2. Perform profile likelihood analysis

   python BO4IO_Post_PL.py -nexp 50 -nrxn 3 -ntrial 5 -niter 250 -ninit 5 -n_data_seed 1 -init_seed 1 -n_p 10 -sigma 0.1

Steps to run the standard pooling problem case study

This case study aims to learn the market demand in the standard pooling problem. All codes should be executed under the standard_pooling folder.

1. Run BO4IO under specific problem settings

   python BO4IO_Std_Pooling.py -nexp 50 -niter 100 -case_study haverly1 -ntrial 5 -sigma 0.05 -theta_seed 1 -n_data_seed 1 -mp 1 -n_p 10 -ntheta 2

   Additional program inputs:

   • case_study - name of benchmark problems, including haverly1, foulds2, and foulds3
   • theta_seed - the starting seed of the random instance, same as init_seed
   • ntheta - number of unknown parameters

2. Perform profile likelihood analysis

   python BO4IO_PostPL.py -nexp 50 -niter 100 -case_study haverly1 -ntrial 5 -sigma 0.05 -theta_seed 1 -n_data_seed 1 -mp 1 -n_p 10 -ntheta 2

3. Perform sensitivity analysis

   python BO4IO_Std_Pooling_sensitivity_analysis.py -case_study haverly1 -ntheta 2 -nexp 50 -mp 1 -n_p 10 -theta_seed 1 -n_data_seed 1 -sigma 0.05

   Additional program inputs:

   nsteps - number of equally-spaced points where the sensitivity analysis perform

Steps to run the generalized pooling problem case study

This case study aims to learn the product quality requirements in the generalized pooling problem. All codes should be executed under the standard_pooling folder.

1. Run BO4IO under specific problem settings

   python BO4IO_Gen_Pooling.py -nexp 50 -niter 200 -case_study Lee -ntrial 5 -sigma 0.05 -theta_seed 1 -n_data_seed 1 -mp 1 -n_p 10

2. Perform sensitivity analysis

   python BO4IO_Gen_Pooling_sensitivity_analysis.py -case_study Lee -nexp 50 -mp 1 -n_p 10 -theta_seed 1 -n_data_seed 1 -sigma 0.05