Code for NeurIPS 2024 paper "Interventionally consistent surrogates for complex simulation models"

Install

Start and activate a new venv with python3.10, navigate to the top directory for this repo, and do python3.10 -m pip install ./

Instructions for running code

Navigate to neurips_ics4csm/ and issue (variables explained below)

python job_script.py --family FAMILY --seed SEEDS --dirname DIRNAME

or

python job_script_bwsg.py --family FAMILY --seed SEEDS --dirname DIRNAME

In the above, set

• FAMILY to one of lode, lodernn, lrnn
• SEED to any sequence of seeds (e.g., 0 1 3 to run code three times at seed 0, 1, and 3)
• DIRNAME to the folder in which you'd like to dump the output of the script

Omitting the --seeds argument will make the seeds default to list(range(5)), which is what was used to generate the 5-fold cross-validation results reported in the paper.

Citation

@inproceedings{dyer2024a,
  publisher = {Neural Information Processing Systems Foundation},
  title = {Interventionally consistent surrogates for complex simulation models},
  author = {Dyer, J and Bishop, N and Felekis, Y and Zennaro, FM and Calinescu, A and Damoulas, T and Wooldridge, M},
  year = {2024},
  organizer = {Neural Information Processing Systems (NeurIPS 2024)}
}