[BUG] type error when running node2vec_embed

[diane-lee-01]

Expected Behavior

Generate a node2vec embedding of the input graph

Actual Behavior

TypeError: unsupported operand type(s) for +: 'int' and 'NoneType' occurs during the _preprocess_transition_probabilities() step in node2vec_embed. I'm not sure why, but unnormalized_probs is currently a list of all 'None's.

Example Code

g = nx.erdos_renyi_graph(10, 0.7, directed=False)
tensors, labels = node2vec_embed(g, dimensions=128, num_walks=10, window_size=2, iterations=3)

The parameters are from layouts\auto.py

Full Traceback

 File "C:\Users\d0156\OneDrive\Documents\GitHub\graspologic\graspologic\layouts\test_n2v.py", line 5, in <module>
    tensors, labels = node2vec_embed(g)
  File "C:\Users\d0156\OneDrive\Documents\GitHub\graspologic\graspologic\embed\n2v.py", line 116, in node2vec_embed
    node2vec_graph._preprocess_transition_probabilities()
  File "C:\Users\d0156\OneDrive\Documents\GitHub\graspologic\graspologic\embed\n2v.py", line 437, in _preprocess_transition_probabilities
    norm_const = sum(unnormalized_probs)
TypeError: unsupported operand type(s) for +: 'int' and 'NoneType'

Your Environment

• Python version: 3.9

Additional Details

Same error occurs for directed graphs too.

[bdpedigo]

@nyecarr any ideas?

[nicaurvi]

Yup, this n2v impl stumbled on unweighted graphs and doesn't really give a good error message. I have a PR that will default weights to 1 and raise a warning for unweighted graphs.

[nicaurvi]

In the meantime, if you are still blocked, you can do something like:

for s, t in g.edges():
    g.add_edge(s, t, weight=1)  # re-add each edge with weight=1

[diane-lee-01]

@Nyecarr Thanks so much, that worked!

I think Word2Vec might have changed the names of some of their arguments though, because I got this error as well:

  File "C:\Users\d0156\OneDrive\Documents\GitHub\graspologic\graspologic\layouts\test_n2v.py", line 7, in <module>
    tensors, labels = node2vec_embed(g,dimensions=128,num_walks=10,window_size=2,iterations=3)
  File "C:\Users\d0156\OneDrive\Documents\GitHub\graspologic\graspologic\embed\n2v.py", line 124, in node2vec_embed
    model = _learn_embeddings(
  File "C:\Users\d0156\OneDrive\Documents\GitHub\graspologic\graspologic\embed\n2v.py", line 205, in _learn_embeddings
    model = Word2Vec(
TypeError: __init__() got an unexpected keyword argument 'size'

And so I changed the arguments 'size' to 'vector_size' and 'iter' to 'epochs' (according to their new documentation) when I was running it. Super minor, but still wanted to let you know.

Again, thanks so much!

[bdpedigo]

i believe that was the reason for pinning gensim's version https://github.com/microsoft/graspologic/blob/5ed36484d32a6ed79db1b7ea087d93fc238df613/setup.cfg#L29

what version of gensim are you running @dlee0156?

[diane-lee-01]

Ohhh I see, I just downloaded the most recent version which was 4.0.1. Thank you!