1. Introduction

This is the official repository for the paper "Beyond Histogram Comparison: Distribution-Aware Simple-Path Graph Kernels". This repository contains the code for reproducing the experiments in the paper.

2. File Structure

The repository is structured as follows:

• datasets/: Contains the graph datasets used in the experiments. All the datasets in paper can be downloaded from here.
• transformers/: save the trained BERT model and the log.
• DASP.py: The implementation of the DASP algorithm.
• DASP_BERT.py: The implementation of the DASP-BERT algorithm.
• pmd.py: The implementation of the probability minkowski distance.
• simple_path_tree.py: The implementation of generating the simple path tree encoding.
• utils.py: The implementation of the utility functions.

3. Requirements

The code is written in Python 3.8.18. The main requirement packages are:

• numpy==1.23.5
• torch==1.11.0
• networkx==2.6.3
• igraph==0.10.8
• transformers==4.35.2
• datasets=2.15.0
• tokenizers==0.15.0
• scikit-learn==1.3.0
• tqdm==4.63.0
• gensim==4.0.1
• joblib==1.4.0
• pandas==1.2.4
• scipy==1.8.0
• pot==0.9.2

And all the running package environments are listed in the environment.yaml file. You can build the environment by running the following command; note that the packages are redundant. We recommend that you install the above packages manually.

conda env create -f environment.yaml

4. Usage

4.1. DASP

To run the DASP algorithm, you can use the following command:

python DASP.py --dataset MUTAG --K 3 --H 2 --size 16

This command will run the DASP algorithm on the MUTAG dataset with the parameters K=3 and H=2, and set the embedding size for word2vec to 16.
Other parameters are set to the default values, which can be found in the DASP.py file.

4.2. DASP-BERT

To run the DASP-BERT algorithm, you can use the following command:

python DASP_BERT.py --dataset MUTAG --K 3 --H 2 --save_model

This command will run the DASP-BERT algorithm on the MUTAG dataset with the parameters K=3 and H=2, and save the trained BERT model. Other parameters are set to the default values, which can be found in the DASP_BERT.py file.