pld-accountant 0.11

PLD accountant

PLD-Accountant

Python code for computing exact DP-guarantees for the subsampled Gaussian mechanism.

The method is described in:

Antti Koskela, Joonas Jälkö, Antti Honkela:
Computing Exact Guarantees for Differential Privacy
https://arxiv.org/abs/1906.03049

Usage

You can download a PyPI package

pip3 install pld-accountant

and run, for example,

from pld_accountant import compute_eps,compute_delta

q=0.01
sigma=1.2
nc=1000 #number of compositions

delta=1e-5

a  = compute_eps.get_epsilon_bounded(q=q,sigma=sigma,target_delta=delta,ncomp=nc)

eps=2.0

d  = compute_delta.get_delta_bounded(q=q,sigma=sigma,target_eps=eps,ncomp=nc)