lejohnyjohn / learning-stance-embeddings-from-signed-social-graphs Goto Github PK

Hi, thanks for sharing the code of your work. I'd love to run this project and see it in action.

I have loaded the BirdwatchSG dataset from HuggingFace:

from datasets import load_dataset
edge_data = load_dataset("Twitter/SignedGraphs")
edge_data = pd.DataFrame(edge_data['train'])

However, I am facing the following error in the last training cell : ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (55,) + inhomogeneous part.
in this part of the code:

Cell In[26], line 124, in TrainingSamples.convert(self, walks, topic_idx, workers)
    122 print("check [len(w) for w in walks]")
    123 print([len(w) for w in walks])
--> 124 walks_lists = np.array_split(walks, workers)

which indicates that the elements in the walks array don't have the same shape.
I have printed the walks array and indeed this is the case:
[2, 2, 3, 2, 2, 3, 2, 2, 3, 2, 3, 2, 2, 2, 3, 3, 2, 3, 2, 2, 3, 3, 2, 2, 3, 2, 2, 2, 2, 4, 3, 2, 2, 2, 3, 2, 4, 2, 3, 2, 3, 2, 3, 3, 3, 3, 3, 2, 2, 3, 3, 2, 2, 3, 2]

This is happening because for some of the walks, there are more neighbors present, so there are longer walk_options :

walk_options = d_graph[walk[-1]].get(neighbors_key, None)

For example for source node: 2665, the walk options are: [482, 2804], and for 482 the walk options are [1595]. So the walk length becomes [2665, 482, 1595], (whereas the walk length is normally 2 for other nodes). But for some source nodes, the walk length is only [2345, 593].

I'd like to know if we need to limit the walk length to be homogenous (i.e., either through padding the length)? or allow for inhomogenous walks?