Unexpected behavior of data.to_heterogeneous()

[DeekshaDixit15]

Hi,

data.to_heterogeneous() function is converting my undirected graph object data to a directed one, any suggestion on how I can prevent it from happening?

[rusty1s]

Do you have a small example to reproduce?

[DeekshaDixit15]

Hi,

Here is an example:

import networkx as nx
from torch_geometric.utils.convert import from_networkx, to_networkx
from random import choices
import torch
import numpy as np

G = nx.complete_graph(100)
node_types = choices([0, 1], k=G.number_of_nodes())
for n in range(G.number_of_nodes()):
G.nodes[n]['node_types'] = node_types[n]
attr = {}
invalid_edges = []
reverse_edges = []
for u,v,a in G.edges(data=True):
edge_types = G.nodes[u]['node_types']+G.nodes[v]['node_types']
if edge_types == 2:
invalid_edges.append((u,v))
else:
attr[(u,v)] = edge_types

G.remove_edges_from(invalid_edges)
nx.set_edge_attributes(G, attr, "edge_types")
print(nx.is_directed(G))
data = from_networkx(G)
data.node_types = np.unique(np.array(data.node_types), return_inverse=True)[1]
data.node_types = torch.from_numpy(data.node_types)

data.edge_types = np.unique(np.array(data.edge_types), return_inverse=True)[1]
data.edge_types = torch.from_numpy(data.edge_types)
print(data.is_directed())
G_data = data.to_heterogeneous(node_type= data.node_types, edge_type=data.edge_types,
node_type_names=['0', '1'],
edge_type_names=[('0', '0', '0'),
('0', '1', '1'), ('1', '1', '0')]
)
print(G_data.is_directed())

[rusty1s]

I think you need to discriminate the edge_type according to the value of G.nodes[u]['node_types']. For example, attr[(u,v)] = 1 if G.nodes[u]['node_types'] == 0andG.nodes[v]['node_types']andattr[(u,v) = 2ifG.nodes[u]['node_types'] == 1andG.nodes[v]['node_types'] == 0`. Otherwise, you are encoding the reverse direction with the same edge type. Let me know!