The Causal Chambers: Experiment Repository

This repository contains the code to reproduce the case studies of the 2024 paper The Causal Chambers: Real Physical Systems as a Testbed for AI Methodology by Juan L. Gamella, Jonas Peters and Peter Bühlmann.

For the datasets, mechanistic models, datasheets and resources to build the chambers, please visit causalchamber.org.

Repository Structure

The code is contained in well-documented Jupyter notebooks, which make use of the causalchamber package to load the pertaining datasets. The following table serves as an index.

Path	Description
case_studies/causal_discovery_iid.ipynb	Case study for causal discovery from i.i.d. data (tasks a1 and a2 in Fig. 5)
case_studies/causal_discovery_time.ipynb	Case study for causal discovery from time-series data (task a3 in Fig. 5)
case_studies/ood_sensors.ipynb	Task b1 (sensor measurements) of the OOD-generalization case study. (Fig. 5b)
case_studies/ood_images.ipynb	Task b2 (images) of the OOD-generalization case study. (Fig. 5b)
case_studies/ood_impulses.ipynb	Task b3 (impulse-response curves) of the OOD-generalization case study. (Fig. 5b)
case_studies/changepoints.ipynb	Change point detection case study (Fig. 5c)
case_studies/ica.ipynb	Case study for independent component analysis (tasks d1, d2 and d3 in Fig. 6)
case_studies/symbolic_regression.ipynb	Symbolic regression case study (Fig. 6e)
case_studies/mechanistic_models.ipynb	Contains examples of running the mechanistic models described in appendix IV, including the code for the case study in (Fig. 6f) and the additional plots in appendix IV.
plots_appendices.ipynb	Contains the code to produce figures 7-15 in appendix III and compute the calibrated reference voltages in Table 3.
causal_validation.ipynb	Contains the code to produce Tables 5-8 in appendix V of the manuscript.

The directory case_studies/src contains supporting code for the notebooks, such as a wrapper for UT-IGSP^[1] and the code for the symbolic regression algorithm^[2] used in the corresponding case study.

Running the notebooks

The Python dependencies to run the notebooks in this repository are specified in the file requirements.txt. We give instructions on how to install these dependencies and run the notebooks using a Python virtual environment, but you may also use other tools such as conda. We ran our experiments using python=3.8.10, but they should work with later versions.

First, create and activate the virtual environment:

python3 -m venv notebooks
source notebooks/bin/activate

Upgrade pip and setuptools, and install the dependencies:

pip install --upgrade pip setuptools
pip install -r requirements.txt

To run the notebooks from the virtual environment, create a new local kernel (while the environment is active):

ipython kernel install --user --name=notebooks

and once inside the notebook select the kernel: Kernel -> Change kernel -> notebooks. You should now be able to run the notebooks in this repository.

Licenses

All code is shared under the MIT license. A copy of the license can also be found in LICENSE.txt.

References

[1] Squires, Chandler, Yuhao Wang, and Caroline Uhler. "Permutation-based causal structure learning with unknown intervention targets." Conference on Uncertainty in Artificial Intelligence. PMLR, 2020.

[2] Kamienny, Pierre-Alexandre, et al. "End-to-end symbolic regression with transformers." Advances in Neural Information Processing Systems 35 (2022): 10269-10281.