This is revisiting and revamping code from an old research project I did while participating in the DIMACS Summer 2015 REU. The goal of the project was to numerically investigate the set of Miller-Rabin witnesses for a large range of numbers, propose some hypotheses derived from analyzing the numerics, and then finally investigate any hypotheses from a purely mathematical standpoint.
You can find a more detailed writeup in the project repository here. If you are interested in the theoretical side of the project, check out the preprint on arxiv.