dp-accounting Documentation

Overview

dp-accounting is a Python library for (differential) privacy accounting. It defines a number of DpEvents, which can be used to describe a variety of mechanisms. These events can be passed to a PrivacyAccountant, which can then compute the privacy parameters for the (composition of) events it has seen so far. For example, this is how we can compute epsilon at delta=1e-5 for a single Gaussian mechanism with noise multiplier 2.0:

accountant = dp.accounting.pld.PLDAccountant()
accountant.compose(dp_accounting.GaussianDpEvent(2.0))
accountant.get_epsilon(1e-5)

Installation

pip install dp-accounting

API Documentation

• API Reference
  • Public API