Closed maberrospi closed 1 week ago
Hi, thanks for using our toolkit. By looking at your error message, can you please check if your K
is all-zero?
Hi, thanks for your quick response!
K using edge weights returned from edge_to_sparse function (following documentation):
K using custom edge features:
K using custom edge features zoomed in:
It seems in both cases K is sparse but is not zero everywhere. Note that both cases result in the same error.
Another possibility might be the values in your K matrix is too large. Please try normalizing them to 0-1
I normalized K (min-max norm) and the error disappears in most cases. I have a case where I still get it but I suppose it might be an error of mine. Thank you for your help!
Good to know! No worries.
I am closing this issue and please feel free to reopen it if the error still exists.
Description: I am trying to use spectral matching to get correspondence between nodes in graphs of different sizes. As per the documentation, I create sparse adjacency matrices, followed by an affinity matrix, which is then passed to the spectral matching function. I get 3 RunTimeWarnings as follows:
To Reproduce Unfortunately I don't have an exact example to add with data but I will provide the bare bones.
Environment:
Additional context FYI. The sinkhorn algorithm column and row sums return NaN. I am sorry I could not provide reproducible code and hope this is enough. Looking forward to your reply.