GCN with Global DP

Description

Code accompanying "Privacy-Preserving Graph Convolutional Networks for Text Classification" paper, run with python GCN.py. Configurations to be specified in settings.py.

Setup

$ sudo apt-get install python3-dev

$ pip install torch==1.6.0+cpu --find-links https://download.pytorch.org/whl/torch_stable.html
$ pip install -r requirements.txt

Run

Experiments can be run with ./run.sh, by default specified for the Cora dataset. Other datasets and parameters can be modified in this bash script, including alternating between private and non-private settings.

An additional script is included for the Pokec dataset, run_pokec.sh, due to additional download and preprocessing steps.

By default, results and other output information will be saved in ./out, while datasets saved to ./data. This can be changed in settings.py, in the out_dir and root_dir variables, respectively.