When we think of driving in general, there are good drivers and bad drivers. So, on a 2D spectrum, we would picture a cluster of data of those drivers and realise that the good drivers’ data is clustered around a particular coordinate(x,y) while the bad drivers’ data is all over the place.
The authors of the paper (Jaewoo Lee et al.) behind this code tutorial proposed bounds for epsilon so that its value may not yield a random output query result that leads to a posterior that is greater than the disclosure risk. In this post, we code their solution.