dp-merf

Dependencies

Versions numbers are based on our system and may not need to be exact matches.

python 3.6
torch 1.3.1              
torchvision 0.4.2
numpy 1.16.4
scipy 1.3.1
pandas 1.0.1
scikit-learn 0.21.2
matplotlib 3.1.0 (plotting)
seaborn 0.10.0 (more plotting)
sdgym 0.1.0 (handling tabular datasets)
autodp 0.1 (privacy analysis)
backpack-for-pytorch 1.0.1 (efficient DP-SGD for DP-MERF+AE)
tensorboardX 1.7 (some logging)
tensorflow-gpu 1.14.0 (DP-CGAN)

Repository Structure

Tabular data

code_tab/single_generator_priv_all.py contains the code for the tabular experiments

Balanced Datasets

code_balanced contains code for MNIST and Gaussian data experiments. See code_balanced/README.md for instructions on how to run the experiments.

comparison models

• dpcgan contains code from https://github.com/reihaneh-torkzadehmahani/DP-CGAN with changes to data-loading
• dpgan contains our implementation of a naive DP-GAN as described in the paper
• gs-wgan contains code from https://github.com/DingfanChen/GS-WGAN with edits to data-loading