Modify the original adjacency matrix

[juanfer]

Hello, I have done modifications to the adjacency matrix such as augmenting by applying several techniques. Is there a way I could update/modify the current g.adj() that comes by default in datasets such as Cora and Citeseer? I understand it works through the edges already created but in this case are the values that changes and not the connections between nodes. Thank you in advance for the help.

[VoVAllen]

@juanfer

I understand it works through the edges already created but in this case are the values that changes and not the connections between nodes.

Hi,

I don’t understand this part you mentioned. Could you elaborate your case? Basically the adj matrix is consistent with the graph edges

[juanfer]

Hi Allen,

Of course, I found a function called “khop_adj” which basically returns the adjacency matrix of a graph to the power of “k”. In my case I want to applied Geometric Scattering to the adjacency matrix by following a wavelet scheme with formulas as:

P = 0.5*(Identity + Adj)
adj_power = P
for _ in k:
adj = torch.spmm(P, adj_power)
return adj

(I have reduced the code for easy understanding). So, it is raising the matrix P by multiplying k times by itself and then return the adjacency matrix of the graph. I basically want to work with the adjacency matrix only. Saying that, I want to save the last matrix in the graph G data. Is that possible? Should it be done considering the edges? How the edges could play a role in this? Thank you for your quick help Allen!