Supporting code for:
A method to challenge symmetries in data with self-supervised learning

• JINST: https://doi.org/10.1088/1748-0221/17/08/P08024
• arXiv: https://arxiv.org/abs/2111.05442
• Zenodo: https://doi.org/10.5281/zenodo.6861702

Test symmetries with sklearn decision tree models.

Setup

Begin from an environment with a recent version of python 3.
We recommend using anaconda:

conda create -n which-is-real python==3.10.4

followed by

conda activate which-is-real

Install and activate a python virtual environment with our required packages:

source setup.sh

You can leave the environment with deactivate.
To clean up fully, remove the installation directory env/.

Run experiments

We include scripts to reproduce all experiments from the paper and a few others.
To run all, execute:

make

This dumps plots into the various directories example_*/.

We describe each in a little more detail below.

Experiment 1: Gappy detector

The cylindrical particle detector with holes and varying efficiencies.
In the paper, this is illustrated in Figures 2 and 3, and discussed in Section 5.

Notebook version: example_ring_paper.ipynb.
You should be able to view this on GitHub with all figures and other outputs.

Execute the plain python version with:

make example_ring_paper.log

This reproduces plots from the paper and dumps them into the directory example_ring_paper/. \

Experiment 2: Height map

The topographical map from Figure 4 and Section 6 of the paper.

Notebook version: example_map_paper.ipynb.
You should be able to view this on GitHub with all figures and other outputs.

Execute the plain python version with:

make example_map_paper.log

This reproduces plots from the paper and dumps them into the directory example_map_paper/.

Bonus experiments:

• make example_ring.log
  • Similar to Experiment 1, plotted in 2D, showing both real and fake (transformed) data.
  • Plots dump to example_ring/.
• make example_map.log
  • Similar to Experiment 2, with more variety in its filtering symmetry-breaking wave.
  • Plots dump to example_map/.
• make example_step.log
  • A minimal example testing translation symmetry on the unit interval.
  • Uses rejection sampling for fakes.
  • Plots dump to example_step/.

These make commands run corresponding example_*.py scripts after ensuring that input data are prepared.