I would like to build GNN model using PyTorch Geometric.
I originally tried to solve the problem of graphs with variable no of node features i.e., the matrix size of I/O data is different using dynamic graph, but when I researched more, it seems that the purpose of dynamic graph is not to learn matrices of different sizes flexibly.
In other words, suppose a full data with a matrix of edge nodes. In this full data, each edge node is represented as 1 if it is connected or 0 if it is not connected.
And then we have data for training purposes.
All data is taken from full data. Full data is embedded as 19700 x 19700 x 50, float 32.
Input data
A: 30 x 30 matrix
B: 500 x 500 matrix
C: 3 x 3 matrix
Output data
G: 24 x 24 matrix
N: 18 x 18 matrix
D: 94 x 94 matrix
I would like to ask if there exists any package/library that's appropriate to build GNN to learn matrices of different sizes as mentioned above?
I'm not a major and would appreciate it if you could point out any inappropriate parts such as terminology.
I used to use dynamic graph, but I found out while working that the tool is not intended for that purpose. So I was wondering if there is another way to do it.