README.md

Differentially private for binary dataset

This repository is for the project of CS4257: Algorithmic Foundations of Privacy

Method:

The basic idea is to generate the synthetic data for differentially privacy based on the marginal tables

Steps:

• Generate the marginal tables: run main.py

  The result will be saved in data/marginal_table_no_noise.pickle

  process = 'generate_marginal_table'
  num_attribute = 600 #number of attributes in the original dataset
  num_marginal_tables = 300 # how many marginal tables
  size_marginal_tables = 8 # how many attributes in one marginal tables
  epsilon = 3 # the differentially private parameter

• Generate the synthetic data: run main.py

  The data will be save in data/result.csv

  num_generate_data = 100 # how many data need to be generated
  num_attribute = 600 #number of attributes in the original dataset
  process = 'generate_data'

  Dataset:

  Please put the dataset in the data folder.