Graph hopper kernel unable to transfom the data

[SantaAnkit]

iam getting a following error while transforming the data in case of graph hopper kernel only.

dataset_name = "MUTAG" # Replace with the desired dataset name
dataset = fetch_dataset(dataset_name, verbose=True)
G = dataset['data'] # Graphs
y = dataset['target'] # Labels
G_train, G_test, y_train, y_test = train_test_split(G, y, test_size=0.2, random_state=42)
gk = GraphHopper(normalize=True,kernel_type='bridge')
'''graph hopper kernel instance'''
K_train = gk.fit_transform(G_train)

Extracting dataset MUTAG..
Parsing dataset MUTAG..
Parse was succesful..
Deleting unzipped dataset files..
Going back to the original directory..

---

ValueError Traceback (most recent call last)
Cell In[98], line 8
6 gk = GraphHopper(normalize=True,kernel_type='bridge')
7 '''graph hopper kernel instance'''
----> 8 K_train = gk.fit_transform(G_train)

File ~/anaconda3/lib/python3.9/site-packages/grakel/kernels/kernel.py:198, in Kernel.fit_transform(self, X, y)
195 self.fit(X)
197 # Transform - calculate kernel matrix
--> 198 km = self._calculate_kernel_matrix()
200 self._X_diag = np.diagonal(km)
201 if self.normalize:

File ~/anaconda3/lib/python3.9/site-packages/grakel/kernels/kernel.py:232, in Kernel._calculate_kernel_matrix(self, Y)
230 cache = list()
231 for (i, x) in enumerate(self.X):
--> 232 K[i, i] = self.pairwise_operation(x, x)
233 for (j, y) in enumerate(cache):
234 K[j, i] = self.pairwise_operation(y, x)

File ~/anaconda3/lib/python3.9/site-packages/grakel/kernels/graph_hopper.py:261, in GraphHopper.pairwise_operation(self, x, y)
258 elif y[0].shape[1] > m:
259 yp = yp[:, :m, :][:, :, :m]
--> 261 return self.metric_((xp.reshape(xp.shape[0], m_sq),) + x[1:],
262 (yp.reshape(yp.shape[0], m_sq),) + y[1:])

File ~/anaconda3/lib/python3.9/site-packages/grakel/kernels/graph_hopper.py:335, in bridge_kernel(x, y)
333 nodepair = (4-NA_i_NA_j_distances)/4
334 nodepair[nodepair < 0] = 0
--> 335 return np.dot(weight_matrix.flat, nodepair.flat)

File <array_function internals>:200, in dot(*args, **kwargs)

ValueError: shapes (196,) and (0,) not aligned: 196 (dim 0) != 0 (dim 0)

I have tried this with other kernel like random walk, shortest path and WL but there it is working.

[giannisnik]

Hi @SantaAnkit ,

I guess that this is because the GraphHopper kernel expects the nodes of the input graphs to be annotated with continuous features, while the nodes of the graphs contained in MUTAG are annotated with discrete labels. This is why no error is thrown for the rest of the kernels you mentioned since those kernels expect discrete node labels.