GLSearch

This repository is the PyTorch implementation of "GLSearch: Maximum Common Subgraph Detection via Learning to Search" (ICML2021). For more details, please refer to our paper.

Installation

Package requirements are in requirements.txt. Please install required packages and versions. We now provide a Dockerfile to make this setup easier.

Mark /model/OurMCS/ and /src/ as the source directories.

Docker Setup

To build and run the Docker environment use the commands below:

cd /path/to/GLSearch
nvidia-docker build -t "glsearch" .
nvidia-docker run -e "HOSTNAME=$(cat /etc/hostname)" -v /path/to/GLSearch:/workspace -it glsearch bash

Datasets

Get the datasets from: https://drive.google.com/drive/folders/1l7_WmO54gPHZ4YwzOqewZtZhWIxn44ZV?usp=sharing

Place the *.klepto files under a new directory /save/OurModelData/ (ex. /save/OurModelData/aids700nef_..._None.klepto).

Google Drive Filename Issues

We noticed Google Drive sometimes renames downloaded files, which will cause issues in data loading. Please check the filenames after downloading.

Data loading is done on line 310 of model/OurMCS/data_model.py (as of commit 4e10d5f79e7f5bd53e74628f52619a0fc10f4c80). You can check here whether tp is pointing to the correct file (NOTE: it is optional whether we include the .klepto extension in tp).

Run

Testing

Select 1 of 3 desired dataset on lines 62-77 of model/OurMCS/config.py by commenting out the rest.

Run the below command:

python3 main.py

Training

Replace config.py with config_training.py.