This repository contains the experimental code and data for the paper: Learning Big Logical Rules by Joining Small Rules, by Céline Hocquette, Andreas Niskanen, Rolf Morel, Matti Järvisalo, and Andrew Cropper, accepted at IJCAI2024.
The results are in the folder /results/{task}/{timeout}/{system}/{trial}.
Joiner is in the folder joiner. You can execute Joiner by runing the command python joiner/popper.py joiner/examples/examples/{ex}
. where ex is the name of an example folder. You can use the flag --datalog --bkcons
for tasks which have a Datalog background knowledge to improve performance.
You can reproduce Figure 5 with the command: python plot_results.py
.
You can reproduce the Figures 3, 4, 6 and 7 with the command: python scatter_plot_results.py
.
You can reproduce the Tables 2 to 5 in the appendix with the command: python table_results.py
.
You can run the learning/training again with the command: python ilpexp.py {experiment_name}
where experiment_name is the name of an experiment in /ilpexp/experiment.py