Implementation of a Graph Attention Network architecture on the PPI dataset.
This repository was made in an attempt to replicate the results presented in the Graph Attention Networks paper on the same dataset. The final result of this implemented presented a F1-score of around 0.90 after around 3000 epochs of training (compared to 0.97 in the paper).
The core structure for the notebook behind this work were provided by the professors behind the Deep Learning in Practice/Advanced Deep Learning class offered in the MSc of Data Sciences and Business Analytics by CentraleSupélec and ESSEC Business School.