chrsmrrs / hashgraphkernel Goto Github PK

Hey,
executing wl.weisfeiler_lehman_subtree_kernel with use_gram_matrix = True and kernel_parameters_wl = [3, True, False, 0] gives the following error.

Traceback (most recent call last):
  File "hash_graph_kernels.py", line 156, in <module>
    main()
  File "hash_graph_kernels.py", line 130, in main
    scale_attributes=True, lsh_bin_width=1.0, sigma=1.0, use_gram_matrices=True)
  File "/media/sf_SEML/code/examples/hashgraphkernel/graphkernel/hash_graph_kernel.py", line 52, in hash_graph_kernel
    feature_vectors = feature_vectors.tocsr()
AttributeError: 'numpy.ndarray' object has no attribute 'tocsr'

By using feature_vectors = sparse.csr_matrix(feature_vectors) this works out.
@chrsmrrs , what do you think?

Thanks for your outstanding job! I am trying to use this HashGraphKernel to calculate similarities between a large number of graphs. However, the number of graphs is too large (about 20,000) to be processed (it will be killed because of out of memory). I want to separate them to avoid out of memory, but it seems some operations have to be taken for all graphs at the same time. Do you have any ideas to solve this problem?

Hi, I just read your paper and try to use your code. As I am not familiar with this area, I got many questions. In the code, the gram matrix is computed for downstream tasks. So I wonder whether the gram matrix is the implicit representation of graphs or explicit representation. It seems the similarity can be calculated by the representation of graphs in each line and also the elements in the matrix already represent their similarity

By the way, your work is very interesting and useful. And I learn a lot from it!