privgem

Privacy-Preserving Generative Models

• ⚠️ Credits
• 🏗️ Installation and setup
• 🧑‍🎓 Tutorials are organized in Jupyter Notebooks as follows:
  • Tabular data
    • PATE-CTGAN
    • DP-CTGAN
    • CTGAN
    • PGM and PATE-CTGAN, using generated/artificial n-class classification problem

Credits

privgem uses codes from other libraries as listed below. ⚠️ Please read the list carefully and cite the original codes/papers as well.

Method	Original version	Notes
PATE-CTGAN	smartnoise-sdk	tabular_patectgan of privgem is based on smartnoise-sdk with some minor changes (e.g., data preproc., logging, plotting and etc).
DP-CTGAN	smartnoise-sdk	tabular_dpctgan of privgem is based on smartnoise-sdk with some minor changes (e.g., data preproc., logging, plotting and etc).
Private-PGM	private-data-generation	tabular_ppgm of privgem is based on private-data-generation with some minor changes.
CTGAN	sdv-dev	currently, privgem uses a forked version of ctgan, link.

Installation

We strongly recommend installation via Anaconda:

• Refer to Anaconda website and follow the instructions.

• Create a new environment for privgem called privgem_py38:

conda create -n privgem_py38 python=3.8

• Activate the environment:

conda activate privgem_py38

• Clone privgem source code:

git clone https://github.com/kasra-hosseini/privgem.git

• Install privgem dependencies:

pip install -r requirements.txt

• Finally, install privgem library:

cd /path/to/privgem
pip install -v -e .

Alternatively:

cd /path/to/privgem
python setup.py install

• To allow the newly created privgem_py38 environment to show up in Jupyter Notebook:

python -m ipykernel install --user --name privgem_py38 --display-name "Python (privgem_py38)"