For this experiment, we use the cora dataset used in the paper "Melding the data-decisions pipeline: Decision-focused learning for combinatorial optimization." by Bryan Wilder, Bistra Dilkina and Milind Tambe.
- Run
get_data.sh
to download preprocessed CORA files - Run
make_cora_dataset.py
to build bipartite graphs and feature vectors associated with each node.
In order to reproduce the result, run the files, whose name start with test
and use the hyperparameter as stated in Appendix B.