Polarization Experiment Simulation

A multi-agent simulation of the polarization process, especially for small groups.

Currently this repository only contains the functions used in the main paper. Some additional functionalities used in the Supplementary section (e.g. the multi-party settings in the "Robustness Test") will be updated later.

main.py The script containing the initialization of one round of simulation, e.g. the agent positions, etc., assigning the parameters to a single run given the user settings, handling the output folder initialization, and calling the simulation.

sim_dyn.py The influence dynamics are implemented in this script. The functions will be called from main.py

sim_utilities.py Containing the polarization metrics. The functions will be called by the other scripts.

The default run of main.py will output a tab-separated log file under the log/current/ folder with the name formats: 1080_3747_T_res_{partyDiff, std} where

• 1080 is the pid-process indentifier number

• 3747 is a random number in (0, 10000) (this is to avoid output clash for multiprocessing)

• {partyDiff, std} indicates the polarization metric outputed.

When the experiment of exo-shock is run, the output is of format:

T  polarization polarization_shock
----------------------------------
0	0.3674382783	0.3792893849

where the first column is the time step, 0 is the first step, the second column is the current polarization metric value, the last column is the polarization on the first dimension of Z. This value is only meaningful after the shock started, which is recorded in the file 1080_3747_T_res_shockRecord.

The code records the polarization change at every 100 steps. But the user could edit the code to increase/decrease the recording interval.

The default run of main.py will output the single run result for $\alpha=.5$, $\beta=.5$, $\gamma=.5$, and $\sigma=.4$ ( $\sigma < .32$ will not be meaningful for the current initialization).

For other parameter settings, please modify the params dictionary list at the top of the __main__ function.

If you want to run multiple experiments on a range of parameters, you could comment out the multiprocessing section below the params dictionary list in the __main__ function within main.py and add your list of parameters in params.

Currently in this vanilla version, no other third party package needs to be installed except for Numpy. Here is the author's environment for a reference:

Python 3.7

numpy=1.18.1

The data, and the code for generating each figure in the main text, as well as the supplementary, could be found in plot/. Please use the .ipynb for generating all except Fig. 5, the heatmap in the main text.

Fig. 5 is generated in Matlab 2018. The data (.mat) and the Matlab scripts (.m) could be found in plot/figure5/.

If you would like to cite this project please use the following citation info:

Macy, M. W., Ma, M., Tabin, D. R., Gao, J., & Szymanski, B. K. (2021). Polarization and tipping points. Proceedings of the National Academy of Sciences, 118(50), e2102144118.