diningphil / E-CGMM

Official Repository of the paper "Modeling Edge Features with Deep Bayesian Graph Networks" (IJCNN 2021)

The Extended Contextual Graph Markov Model (E-CGMM)

a.k.a. making CGMM process arbitrary edge features

Description

Here you will find all you need to replicate the experiments in our code (please look at previous releases).

If you happen to use or modify this code, please remember to cite our paper:

Atzeni Daniele, Bacciu Davide, Errica Federico, Micheli Alessio: Modeling Edge Features with Deep Bayesian Graph Networks, IJCNN, 2021.

Usage

This repo builds upon PyDGN, a framework to easily develop and test new DGNs. See how to construct your dataset and then train your model there.

This repo assumes PyDGN 1.0.3 is used. Compatibility with future versions is not guaranteed, e.g., custom metrics need to be slightly modified starting from PyDGN 1.2.0.

The evaluation is carried out in two steps:

• Generate the unsupervised graph embeddings
• Apply a classifier on top

We designed two separate experiments to avoid recomputing the embeddings each time. First, use the config_CGMM_Embedding.yml config file to create the embeddings, specifying the folder where to store them in the parameter embeddings_folder. Then, use the config_CGMM_Classifier.yml config file to launch the classification experiments.

Launch Exp:

Build dataset and data splits (follow PyDGN tutorial and use the data splits provided there)

For instance:

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

Train the model

pydgn-train  --config-file MODEL_CONFIGS/config_ECGMM_Embedding.yml 
pydgn-train  --config-file MODEL_CONFIGS/config_ECGMM_Classifier.yml