Just adding Laplace and Gaussian noise#

Using higher-level APIs helps to ensure the correctness of your analysis, but sometimes you just want to add noise.

[1]:

import opendp.prelude as dp

dp.enable_features("contrib")

[2]:

space = dp.atom_domain(T=float, nan=False), dp.absolute_distance(T=float)
laplace_mechanism = space >> dp.m.then_laplace(scale=1.0)
dp_value = laplace_mechanism(100.0)
print(dp_value)

98.80781116389129

Ta-da! We’ve added calibrated Laplace noise to 100.0. Gaussian noise is also possible:

[3]:

gaussian_mechanism = space >> dp.m.then_gaussian(scale=1.0)
dp_value = gaussian_mechanism(100.0)
print(dp_value)

99.69869099986073

If you have any doubts about whether Laplace or Gaussian noise is appropriate, or how to choose an appropriate scale, the Context API will handle those details for you.