I am confused about the exact layers in the architecture from the paper. It states that:
"
We use the “mean” variant of GRAPHSAGE [16] and apply a DIFFPOOL layer after every two GRAPHSAGE layers in our architecture. A total of 2 DIFFPOOL layers are used for the datasets. For small datasets such as ENZYMES and COLLAB, 1 DIFFPOOL layer can achieve similar performance. After each DIFFPOOL layer, 3 layers of graph convolutions are performed, before the next DIFFPOOL layer, or the readout layer.
"
So 2 or 3 GraphSAGE layers are used? Which one of the below would be correct? Or if no one is correct, what would be the exact architecture from the paper?
1. GraphSAGE
2. GraphSAGE
3. DIFFPOOL
4. GraphSAGE
5. GraphSAGE
6. DIFFPOOL
7. GraphSAGE
8. GraphSAGE
9. GraphSAGE
10. READOUT
1. GraphSAGE
2. GraphSAGE
3. GraphSAGE
4. DIFFPOOL
5. GraphSAGE
6. GraphSAGE
7. GraphSAGE
8. DIFFPOOL
9. GraphSAGE
10. GraphSAGE
11. GraphSAGE
12. READOUT
1. GraphSAGE
2. GraphSAGE
3. DIFFPOOL
4. GraphSAGE
5. GraphSAGE
6. DIFFPOOL
7. READOUT
I am confused about the exact layers in the architecture from the paper. It states that:
"
We use the “mean” variant of GRAPHSAGE [16] and apply a DIFFPOOL layer after every two GRAPHSAGE layers in our architecture. A total of 2 DIFFPOOL layers are used for the datasets. For small datasets such as ENZYMES and COLLAB, 1 DIFFPOOL layer can achieve similar performance. After each DIFFPOOL layer, 3 layers of graph convolutions are performed, before the next DIFFPOOL layer, or the readout layer.
"
So 2 or 3 GraphSAGE layers are used? Which one of the below would be correct? Or if no one is correct, what would be the exact architecture from the paper?