为什么来自 networkx 的 read_edgelist 不接受数组/列表作为权重？

Question

I am using Python and the networkx package to read an edgelist from file in order to build a graph.

My edgelist looks something like this:

[(0, 114, {'pts': array([[  1, 822],
       [  1, 821],
       [  2, 820],
       [  3, 819]], dtype=int16), 'weight': 23.38477631085024}),
(1, 110, {'pts': array([[ 1,  3],
       [ 1,  2],
       [ 2,  1]], dtype=int16), 'weight': 18.414213562373096})]

I wrote this edgelist with:

nx.write_edgelist(G, 'my_el.edgelist', data=True)

My edges are defined by the start and the end node, and following that, I have my weights. Each edge has two weights. The first weight is a array of pixel-coordinates, and the second is a float.

The graph was constructed from a skeleton with the ´sknw´ library using the build_sknw function :

def build_sknw(ske, multi=False):
    buf = buffer(ske)
    nbs = neighbors(buf.shape)
    acc = np.cumprod((1,)+buf.shape[::-1][:-1])[::-1]
    mark(buf, nbs)
    pts = np.array(np.where(buf.ravel()==2))[0]
    nodes, edges = parse_struc(buf, pts, nbs, acc)
    return build_graph(nodes, edges, multi)

Now I want to read in this edgelist to build a graph. However, Python does not recognize my array of pixel-coords as a single weight-element. I've tried nx.read_edgelist('my_el.edgelist', data=True) , which gives me the following error:

TypeError: Failed to convert edge data (["{'pts':", 'array([[', '1,', '822],']) to dictionary.

nx.read_edgelist('my_el.edgelist', data=['pts', 'weight'] gives me:

IndexError: Edge data ["{'pts':", 'array([[', '1,', '822],'] and data_keys ['pts', 'weight'] are not the same length

and nx.read_edgelist('my_el.edgelist', data=(('pts', int), ('weight', float'))) gives me

IndexError: Edge data ["{'pts':", 'array([[', '1,', '822],'] and data_keys (('pts', <class 'int'>), ('weight', <class 'float'>)) are not the same length

I assume that the function is having a problem with either the array as a weight, or with the formatting of the my_el.edgelist file, but I do not really know, how to properly solve this issue without any workarounds via conversion to string, or similar.

I'd be thankful if someone can point me in the right direction and help me out with this!

Answer 1

Dealing with numpy arrays seems to be a major problem for networkx. The function that converts my image skeleton to the graph G however seems to be using numpy arrays. Since I prefer not to alter the imported function, a possible workaround that seemed to mitigate the problem when writing the edgelist to file was to change the array type thus:

for (s, e) in G.edges():
    G[s][e]['pts'] = G[s][e]['pts'].tolist()

It might not be the most computationally efficient, and doesn't deal with the root of the problem, but will get the job done, in case anybody encounters a similar issue.