I am trying to use the Kolter algorithm provided in the repository. I am not able to run the code on the eco dataset. I get an error in the function SpectralClustering because the first input argument of kmeans is not real. This first input argument are the eigenvectors of the Laplacian L= D-W. Apparently in the setting at hand the adjacency matrix W is not symmetric and, thus, in general D might have complex eigenvectors. If the adjacency matrix is allowed to be non-symmetric, then I think this whole method might have a flaw. Otherwise I assume there is a bug in the computation of the adjacency matrix.
Hi there,
I am trying to use the Kolter algorithm provided in the repository. I am not able to run the code on the eco dataset. I get an error in the function SpectralClustering because the first input argument of kmeans is not real. This first input argument are the eigenvectors of the Laplacian L= D-W. Apparently in the setting at hand the adjacency matrix W is not symmetric and, thus, in general D might have complex eigenvectors. If the adjacency matrix is allowed to be non-symmetric, then I think this whole method might have a flaw. Otherwise I assume there is a bug in the computation of the adjacency matrix.
Best, Andi