salesforce / GAEA

Data and code for Salesforce Research paper, GAEA: Graph Augmentation for Equitable Access via Reinforcement Learning - https://arxiv.org/abs/2012.03900 . The paper provides methods for constraint graph augmentation and optimal facility placement problems

GAEA: Graph Augmentation for Equitable Access via Reinforcement Learning

This repository contains the data and code for Salesforce Research paper: GAEA: Graph Augmentation for Equitable Access via Reinforcement Learning

Citation

If you use this code, data or our results in your research, please cite as appropriate:

@inproceedings{ramachandran2021gaea,
  title={GAEA: Graph Augmentation for Equitable Access via Reinforcement Learning},
  author={Ramachandran, Govardana Sachithanandam and Brugere, Ivan and Varshney, Lav R and Xiong, Caiming},
  booktitle={Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society},
  pages={884--894},
  year={2021}
}

Prerequisites

Software

Install dependencies by running

pip install -r requirements.txt

The code was tested on

cuda 11.2
tensorflow-gpu==2.6.0
keras==2.6.0

Hardware

We ran on a Quadro GV100 with 32GB RAM.

Dataset

1. Dataset merging public census, school, and transportation datasets for the city of Chicago is provided under data/{demographics | network | schools}
2. For Facebook100 dataset download the data as described in http://sociograph.blogspot.com/2011/03/facebook100-data-and-parser-for-it.html and place the unziped data under data/facebook100

Experiments

Edit repository path and the output path for the project in paths_inc.py .

The run_experiments.py generates all results for:

1. Original graph
2. Baseline method
3. Proposed method

On each of the outputted graphs, we run monte carlo weighted walk simulations and estimate the distribution of expected rewards of walkers. On this distribution, we evaluate our main two criteria:

1. Expected Utility
2. Gini Index of Expected Utility

Graph editing on Chicago school network

python run_experiments.py --exp edit --graph chicago

Graph editing on Facebook100 schools

python run_experiments.py --exp edit --graph fb --school Caltech36

Other school network we tried are: Mich67 and Reed98

Graph editing on synthetic network

python run_experiments.py --exp edit --graph synthetic

Facility Placement

python run_experiments.py --exp facility_placement --graph chicago