UnlearnSPN

Installation

1. Create a new conda environment from the conda.yml file:

conda env create -n unlearnspn --file conda.yml

2. Activate the conda environment:

conda activate unlearnspn

3. Unpack the modified version of SPFlow:

unzip SPFlow.zip

4. Navigate to SPFlow source directory and install it manually:

cd SPFlow/src/
python setup.py install

5. Unzip datasets:

unzip data.zip

Reproducing the results

In order to reproduce the results from the paper, simply run:

./run_training_comparison.sh && ./run_experiments.sh

The results will be stored using mlflow. In order to view the results run

mlflow ui

in the directory containing the mlruns directory. This will start a local web server, which can be accessed through 127.0.0.1:5000 in your web browser.