impelement GIN and GCN

[Anshul22Verma]

To implement GIN and GCN for basic datasets in PyTorch.

• [x] GCN -> paper1
• [x] GIN ->paper2
• [ ] reusable code 🔥🔥🔥🔥
  ♻️ 🚀

[Anshul22Verma]

4 added GIN and GCN basic examples will improve the implementation in v2

[Anshul22Verma]

Note

with @aalok1993 @prajwalsingh

---

Q. When to stop WL-algorithm ?

A. @aalok1993, Stop when the pattern stops changing. Say a star graph -> one vertex in the center.

First iteration: Central node will receive information from all the 5 neighbors and neighbor receive information from the central node.

Second iteration: Central node revives the same iteration again from the the neighbors with its initial state.

This can loop forever. So you stop when the pattern stops changing not when the color stops changing.

@aalok1993 to put image.

Example 2

Once the stable coloring is reached then we stop the iteration. --> O(|V| + |E|) iteration where |V| is number of nodes in G1 and |E| is nodes in G2

Bell Number finds the total possible partition variations in a set which we use to maximum number of iterations of a WL-test.

Next Week:

When to stop in 1-dimensional WL-test and when to stop in k-dimensional WL-test. #7