NeuroCUT: A Neural Approach for Robust Graph Partitioning

This repository is an official implementation of the Paper : NeuroCUT: A Neural Approach for Robust Graph Partitioning, accepted in KDD'24.

Commands to run

1. Activate Environment or create from requirements.txt

   source {environment name}/bin/activate

2. Various parameters are listed in makefile, run commmand

   make parameter_1={value_1} parameter_2={value_2} ..

Example command has been provided in the run.sh file.

Reults will be formed in the results folder, and model will be saved in models folder.

Sample Dataset

A sample Cora graph is given in the data folder. The structure of input graph is as follows:

<Graph_name> //input to the model
    <test_set> 
        <1>...<n>
    <val_set> 
        <1>...<n>
    <train_set> 
        <1>...<n>
// Each folder in train/val/test set should have a graph.txt and graph_stats.txt. The number of required cuts can be modifited in the graph_stats file.

Also, to run on any new graph, you need to add the graph.txt and node_embedding.pt file in raw_data folder.

Adding a new dataset

1. Add the graph.txt and node_embedding.pt file in the raw_data folder. You can use the a.py file to download and modify the dataset.
2. Create an experiment in the data folder containing the train, test, and validation subdirectories. You will need to add the graph.txt and graph_stats.txt file there.
3. Our framework supports training on multiple graphs and inferring on different graphs as well. Configure your experiment folder accordingly to take advantage of this flexibility.