Skip to content

Empirical Research over the possible advantages of pretraining a Graph Neural Network for Classification by using Link Prediction. We used GCN, GAT and GraphSAGE with minibatch generation. Done for the Learning From Networks course taught by professor Fabio Vandin at the University of Padova

Notifications You must be signed in to change notification settings

AlbertoFormaggio1/fine_tuning_classification_prediction_GNN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

b71f597 · Feb 12, 2024

History

80 Commits
Jan 25, 2024
Jan 25, 2024
Feb 12, 2024
Jan 24, 2024
Jan 25, 2024
Jan 25, 2024
Nov 19, 2023
Jan 25, 2024
Jan 25, 2024
Jan 21, 2024
Jan 25, 2024
Dec 7, 2023

Repository files navigation

GNN Integration for Node Classification and Link Prediction

Research Overview:

This repository investigates the collaborative dynamics between node classification and link prediction in the realm of Graph Neural Networks (GNNs). The primary goal is to evaluate the potential advantages gained by integrating these tasks within a graph, as opposed to solely utilizing GNNs for node classification. By merging the predictive capabilities of link prediction with the node classification features of GNNs, the aim is to enhance the overall performance and effectiveness of graph-based models.

Implementation Details:

  • In the "main.py" file, the combination of GNN + MLP (Multi-Layer Perceptron) + Link Prediction is implemented.
  • For an alternative approach without MLP, refer to the "main_no_MLP.py" file, utilizing the combination of GNN + Link Prediction.

Configuration:

Before running the code, navigate to the COMMANDS section (lines 25-30) in either "main.py" or "main_no_mlp.py" to set the preferred GNN (GCN, GAT, SAGE) and dataset (cora, citeseer, pubmed). For hyperparameter testing, enable the "use_grid_search" option. Adjust hyperparameters in the "parameters.py" file, filling the "parameters_[GNN]" dictionary for specific sets, or use "parameters_grid_[GNN]" for comprehensive hyperparameter combination testing by setting "use_grid_search = True".

About

Empirical Research over the possible advantages of pretraining a Graph Neural Network for Classification by using Link Prediction. We used GCN, GAT and GraphSAGE with minibatch generation. Done for the Learning From Networks course taught by professor Fabio Vandin at the University of Padova

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages