I looked into the implementation of a graph attention layer in pytorch-geometric. A graph attention network was introduced by Velickovic et al. in their paper "Graph Attention Networks". In this video, the focus is on (1) how pytorch-geometric implemented a graph attention layer (2) the expressive power of GNNs (3) training and testing graph attention on graph datasets.
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
🔗 notes + code: https://mashaan14.github.io/YouTube-channel/graph_neural_networks/2024_02_05_GAT
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
- website https://mashaan14.github.io/mashaan/
- github https://github.com/mashaan14
- X https://twitter.com/mashaan_14
- linkedin https://linkedin.com/in/mashaan
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
- sponsor this channel on GitHub Sponsors:
https://github.com/sponsors/mashaan14
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
📹 Video edit: Adobe Premiere Rush
🎧 Audio enhancement: Adobe Podcast
🖼️ Thumbnails: GIMP
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Chapters:
0:00 start
0:15 reading the paper
1:52 visualizing multi head graph attention
2:43 GATConv layer in pytorch-geometric
3:25 class GATConv
5:00 coding a graph attention layer
5:35 creating a synthetic dataset
6:08 create Data instance for pytorch-geometric
6:41 creating an instance of class GATConv
7:04 train and test functions
7:19 training graph attention
7:57 training an MLP
8:26 how come an MLP beats a GNN?!!
9:06 the expressive power of GNNs
10:18 rule number 1 for using GNNs
10:33 importing CORA dataset
11:28 results summary
11:51 final remarks
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
#graphneuralnetwork #GATConv #pytorch #graph #convolution #GNN #GCN #pytorch #pytorchgeometric #GCNConv #DeepLearningTutorial #MachineLearningProject #AIResearch #CodingTutorial