anonymeter-sdnist

This repository contains the code used to analyze the privacy of de-identified data submitted as part of the NIST Collaborative Reserach Cycle (CRC) using Anonymeter. The results of this analysis will be published soon in the form of a tiny paper.

Setup

To run this you need to clone the CRC repository in this repository.

git clone [email protected]:usnistgov/privacy_collaborative_research_cycle.git

In particular put the crc_data_and_metric_bundle_1.1 directory with all the de-identified data submissions in this directory. Then you also need to download the SDNIST repository with the NIST Diverse Communities Data Excerpts original and control datasets.

Finally, you need to install the required packages, including anonymeter, to be able to run the privacy evaluations:

pip install -r requirements.txt

Structure

There are a few modules in this repository:

• anonymeter_evaluation_lib.py: contains convenience high-level classes to run privacy attacks for different hyperparameter settings.
• analysis.py: this is the main analysis script. It will run everything. It is recommended to provision enough computing power for a more pleasant experience.
• analysis_utils.py: low-level utility functions needed to run the analysis.
• plot_results.py: script to produce all the plot in the paper.
• plot_functions.py and plot_utils.py: low level plotting functions and utilities.