abcdp

Approximate Bayesian Computation with Differential Privacy

How to install?

It is best to use Anaconda. Make sure you create a separate Anaconda environment for this project. Use Python 3.

To install, first switch to the right environment. If you use Anaconda, then you can do this with conda activate name_of_environment. This repository contains a Pyton-importable package that can be installed with pip command:

pip install -e /path/to/the/code/folder/of/this/repo/on/your/local/drive

The "code" folder is the one that contains setup.py. Also do:

pip install autodp

Once installed, you should be able to do import abcdp without any error.

How to run the experiments?

Flip probability plot

• Run ./ABCDP/abcdp/prob_flip.py in order to get Figure 1.

Toy experiment

• Run ./ABCDP/abcdp/auxiliary_files/run_toy_example.py to compute the ABCDP algortihm for different parameter settings.
• To obtain Figure 2, run ./ABCDP/abcdp/figure_toy_example.py.

Coronavirus outbreak data

• To obtain Figure 4 run ./ABCDP/code/abcdp/figure_covid.py

Modelling Tuberculosis (TB) Outbreak

• Run ./ABCDP/abcdp/auxiliary_files/test_TB.py file to get the results.
• For plotting Figure 5 run ./ABCDP/abcdp/figure_covid.pyfile.