Wasserstein Weisfeiler-Lehman Graph Kernels
wwl, run the following:
$ pip install cython numpy $ pip install wwl
WWL can be used to compute the pairwise kernel matrix between a list of Graphs.
The kernel function
wwl takes as input a list of igraph
Graph objects. It can also take their node features (if they are continuously attributed), the number of iterations for the embedding scheme,
the value for gamma in the Laplacian kernel, and a flag for sinkhorn approximations.
from wwl import wwl # load the graphs graphs = [ig.read(fname) for fname in graph_filenames] # load node features for continuous graphs node_features = np.load(path_to_node_features) # compute the kernel kernel_matrix = wwl(graphs, node_features=node_features, num_iterations=4) # use in SVM from sklearn.svm import SVC train_index, test_index = np.load(train_index_path), np.load(test_index_path) y = np.load(path_to_labels) K_train = kernel_matrix[train_index][:,train_index] K_test = kernel_matrix[test_index][:,train_index] svm = SVC(kernel='precomputed') # For a Krein SVM, please refer to krein.py svm.fit(K_train) y_predict = svm.predict(K_test)
utilities.wwl_custom_grid_search_cv for a custom crossvalidation to cross-validate the number of iterations, gammas in the Laplacian kernel, and other parameters for the SVM.
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size wwl-0.1.2-py3-none-any.whl (8.2 kB)||File type Wheel||Python version py3||Upload date||Hashes View hashes|
|Filename, size wwl-0.1.2.tar.gz (7.0 kB)||File type Source||Python version None||Upload date||Hashes View hashes|