Different Graph Neural Network Implementation using PyTorch Geometric

Implement GCN, GraphSAGE, and GAT on PubMed using PyTorch_Geometric

Dataset:

The PubMed dataset consists of scientific publications from the PubMed database on types of diabetes classified into one of three classes

• Diabetes Mellitus Experimental
• Diabetes Mellitus Type_1
• Diabetes Mellitus Type_2

The citation network consists of 44,338 links. Each publication in the dataset is described by a TF/IDF weighted word vector from a dictionary which consists of 500 unique words.

Objective

The goal is to classify each publication into one of the three classes by performing Node classification using GCN, GraphSAGE, and GAT.

t-SNE visualization of PubMed(image by author)

Techniques

GCN(Graph Convolutional Neural Network) is a spatial framework that takes graphs as an input and applies convolution operations over the Graph, similar to CNN, which applies convolutions to the images. GCN is based on graph convolutions built by stacking multiple convolutional layers and a point-wise non-linearity function like ReLu following each layer.