Numpy wrongly converts Networkx graphs to array

Question

I got an issue when trying to cast a networkx graph list into a Numpy array. It appears that when all graphs of the list have the same number of nodes, Numpy automatically converts input graphs into arrays and the result is not what I expect.

Here is a small reproduction script:

import numpy as np
import networkx as nx

g1 = nx.DiGraph([(1, 2), (2, 3)])
g2 = nx.DiGraph([(1, 2), (2, 3), (3, 4)])
g3 = nx.DiGraph([(1, 2), (2, 1), (1, 3)])

ok_numpy = np.array([g1, g2], dtype=object)
ko_numpy = np.array([g1, g3], dtype=object)

>> ok_numpy = [<networkx.classes.digraph.DiGraph object at ...>, <networkx.classes.digraph.DiGraph object at ...>]
>> ko_numpy = [[1 2 3], [1 2 3]] <-- Objects are not DiGraph() anymore, edges info is lost, dimension is 2 instead of 1

If anybody knows a solution so that ko_numpy could behave the same way ok_numpy does, it would be perfect.

Answer 1

This is a bug-feature of numpy, see this answer. The current solution is to create an empty array and assign specific elements to it:

ko_numpy2 = np.empty(2, dtype=object)
ko_numpy2[:] = [g1, g3]

Not ideal, but workable.