Machine Learning project 22/23

Francesco Bottalico (787587)

Importing

Code for the link prediction task on graphs using autoencoders. link_prediction is the python package that can be easily imported by doing

import link_prediction

Notebook

The code in this repository is the same present in the Colab Notebook, which was used to run the experiments (The notebook is in the notebook folder or here).

Dependencies

The following dependencies are needed in order to use the package:

• scikit-learn
• dgl
• torch
• numpy
• scipy

Features

The package exposes the model's classes used during the experiments:

• Graph Convolution AutoEncoder (GCAE)
• Graph Normalized Convolution AutoEncoder (GNCAE)
• Graph ATtention AutoEncoder (GATAE)
• Graph Transformer AutoEncoder (GTAE)

And the main functions to train them using k-fold:

• train_kfold for simple training
• train_contrastive_kfold for contrastive training