HMM4G: a Hidden Markov Model for Temporal Graph Representation Learning

The repository contains the scripts to reproduce the results of the paper

Errica F., Gravina A., Bacciu D., Micheli A., Hidden Markov Models for Temporal Graph Representation Learning, ESANN 2023.

If you found the code and paper useful, please consider citing us:

@inproceedings{errica_hidden_2023,
title={Hidden Markov Models for Temporal Graph Representation Learning},
author={Errica, Federico and Gravina, Alessio and Bacciu, Davide and Micheli, Alessio},
booktitle={Proceedings of the 31st European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN)},
year={2023},
}

How to reproduce

The steps below are used to reproduce both model selection and risk assessment phases of HMM4G.

Step 1) Create venv environment

source create_environment.sh [cpu/cu112/cu113/cu116/cu117]

This will create a venv environment called hmm4g with cpu or cuda support depending on your choice.

We will be using pydgn 1.3.1 to run the experiments.

Step 2) Create dataset

You can prepare the dataset using the following command

pydgn-dataset --config-file DATA_CONFIGS/config_pedalme.yml

(and similarly for the other datasets using the configuration files in the DATA_CONFIGS folder.)

Step 3) Launch node embedding experiment (remove debug mode to parallelize)

pydgn-train --config-file MODEL_CONFIGS/pedalme/config_HMM4G_gaussian_embedding.yml --debug

This will create the HMM4G embeddings in the HMM4G_EMBEDDINGS folder.

(and similarly for the other datasets using the configuration files in the MODEL_CONFIGS folder.)

Step 4) Launch classification task (remove debug mode to parallelize)

pydgn-train --config-file MODEL_CONFIGS/pedalme/config_HMM4G_classifier_mlp.yml --debug

This will launch model selection and risk assessment for the readout MLP and compute the final scores.