Code for: Graph-based Methods for Discrete Choice

This repository accompanies the paper:

Graph-based Methods for Discrete Choice.
Kiran Tomlinson and Austin R. Benson.
Network Science, 2023.

Contents

This repository includes one directory:

• code/: all code needed to process data, implement our methods, run our experiments, and plot results

In addition, the following directories containing our results and data can be downloaded from https://osf.io/egj4q/:

• raw-data/: raw data files for the election data
• data/: processed versions of our datasets
• results/: our experiment result files

The instructions below assume these directories are placed alongside code/.

Data

In https://osf.io/egj4q/, we provide our processed versions of the us-elec-2016, ca-elec-2016, and ca-elec-2020 datasets as well as the original raw data (original sources: https://statewidedatabase.org/election.html, https://github.com/mkearney/presidential_election_county_results_2016, and https://github.com/000Justin000/gnn-residual-correlation/tree/master/datasets/election). In the code, they are named election-2016, ca-election-2016, and ca-election-2020. The Friends and Family datasets are not licensed for redistribution so we have not included them here, but they can be requested from http://realitycommons.media.mit.edu/friendsdataset.html and our data processing code is included (see reproducibility instructions below). Our California data is built from the 2016 and 2020 General Election data SOV (statement of vote) and REG (registration) files, per-county codebooks, and precinct SRPREC_SHP files. These are all available at https://statewidedatabase.org/election.html, but we include them for completeness.

Reproducibility

We provide instructions for replicating all experiments.

1. Setup

Download the raw-data/, data/, and results/ folders from https://osf.io/egj4q/ and place them alongside code/. After requesting the Friends and Family dataset as described above, place the Friends and Family files in raw-data/ (in particular, App.csv, AppRunning.csv, and BluetoothProximity.csv). Then, modify config.yml so that parsed_data_dir contains the path to the data/ directory included in this repo, data_dir contains the path to raw-data/, and results_dir contains the path to the results/.

2. Friends and Family data processing

Simply run python3 parse-data.py to generate the processed app-install and app-usage directories in data/.

3. Experiments

First, set the number of threads you want to run on (we used 100)

export NUM_THREADS=[number of threads to run on]

To run the synthetic data sample complexity experiment:

python3 expriments.py --network-convergence --threads $NUM_THREADS

To run all the election experiments:

python3 experiments.py --election-social election-2016 --threads $NUM_THREADS
python3 experiments.py --election-adjacency ca-election-2016 --threads $NUM_THREADS
python3 experiments.py --election-adjacency ca-election-2020 --threads $NUM_THREADS
python3 experiments.py --election-social election-2016 --propagation --threads $NUM_THREADS
python3 experiments.py --election-adjacency ca-election-2016 --propagation --threads $NUM_THREADS
python3 experiments.py --election-adjacency ca-election-2020 --propagation --threads $NUM_THREADS

To run all the app experiments:

python3 experiments.py --masked --datasets app-usage app-install --per-item-utilities --app-prox-thresholds 10 --threads $NUM_THREADS 
python3 experiments.py --propagation --datasets app-usage app-install --app-prox-thresholds 10 --threads $NUM_THREADS 

To run the runtime experiment:

python3 expriments.py --timing

4. Plotting

Simply run

python3 plot.py

This will print out data for all the tables and save plots to code/plots/.