CODEWITHSUNDEEP
pythonpython-3.xnetworkxmathematical-optimization
Buskruid
Buskruid

Reputation: 63

How to fix inconsistent labelling of edges/nodes in Python's networkx?

I want to define an objective function for the max-cut problem on graphs. I use the following expression, 0.5*sum([w[i,j]*(1-spin[i]*spin[j]) for i,j in G.edges]) where G is a networkx graph, w is the numpy matrix generated from that graph (connectivity matrix), and spin is an array with entries -1 or 1 to denote which side of the partition the node is in.

All good I thought, but it turns out that the labelling of the edges is not consistent with the labelling of the nodes, see code below. For example in the graph G1, edge weight w[1,5] is 0 even though the edge (1,5) is in the graph. Any recommendations on how to fix this?

Cheers

import networkx as nx

G1 = nx.random_regular_graph(3,6, seed = 1)
G2 = nx.random_regular_graph(3,6, seed = 1)

# labelling seems not to be conserved when transforming to matrix and back
G2 = nx.to_numpy_matrix(G2)
G2 = nx.from_numpy_matrix(G2)

print(nx.to_numpy_matrix(G1))
print(G1.edges)
print(nx.to_numpy_matrix(G2))
print(G2.edges)

Output

[[0. 1. 0. 1. 0. 1.]
 [1. 0. 1. 0. 1. 0.]
 [0. 1. 0. 1. 0. 1.]
 [1. 0. 1. 0. 1. 0.]
 [0. 1. 0. 1. 0. 1.]
 [1. 0. 1. 0. 1. 0.]] # matrix of G1

[(0, 1), (0, 4), (0, 3), (1, 2), (1, 5), (2, 3), (2, 4), (4, 5), (5, 3)] # edges of G1

[[0. 1. 0. 1. 0. 1.]
 [1. 0. 1. 0. 1. 0.]
 [0. 1. 0. 1. 0. 1.]
 [1. 0. 1. 0. 1. 0.]
 [0. 1. 0. 1. 0. 1.]
 [1. 0. 1. 0. 1. 0.]] # matrix of G2
[(0, 1), (0, 3), (0, 5), (1, 2), (1, 4), (2, 3), (2, 5), (3, 4), (4, 5)] # edges of G2

Upvotes: 1

Views: 178

Answers (1)

Sparky05
Sparky05

Reputation: 4892

As described in to_numpy_matrix documentation, the method implicitly uses the ordering of G.nodes(), if the parameter nodelist is not used.

Using the following code should fix the ordering

nx.to_numpy_matrix(G, nodelist=sorted(G))

Upvotes: 0

Related Questions