How to get mediate feature of graphs about grakel HOT 1 CLOSED

Hi @LittleWhite-Hai ,

For some kernels where the mapping to the feature space is explicit, you can indeed obtain such features. For instance, for the Weisfeiler-Lehman subtree kernel, you can obtain the features for the different iterations as follows:

from grakel.datasets import fetch_dataset
from grakel.kernels import WeisfeilerLehman, VertexHistogram

dataset = fetch_dataset("MUTAG", verbose=False)
Gs = dataset.data
wl_kernel = WeisfeilerLehman(n_iter=3,  base_graph_kernel=VertexHistogram)
wl_kernel.fit(Gs)
features_it_0 = wl_kernel.X[0].X
features_it_1 = wl_kernel.X[1].X
features_it_2 = wl_kernel.X[2].X
features_it_3 = wl_kernel.X[3].X

In this example, features_it_0, features_it_1, features_it_2, features_it_3 are all sparse matrices where the number of rows is equal to the number of graphs (i.e., len(Gs)) and the number of columns is equal to the number of different labels assigned to the nodes of all graphs at that iteration.

from grakel.