community generation task rethinking

[v18nguye]

Hi authors,

Thank you for your published code!

I have skimmed through the repo. As I am currently interested in the graph generation problem, so I have looked deeper into the Community generation problem. There is an abnormal problem in the code that generates the Community dataset, where a hundred of generated graph samples have the same number of nodes, and the same number of nodes for each generated community given a graph sample. I think that It is not a standard way that researchers benchmark models on this dataset. After fixing this issue by having different number of nodes per community per generated graph, I tested again your model and the results changed significantly, and that does not resemble to the published results.

what I have changed is to put c_sizes in the k loop to have various graph sizes and community sizes.

MGVAE/general_graph_generation/create_graphs.py

Line 66 in d254519

c_sizes = np.random.choice([5, 6, 7, 8, 9, 10], num_communities)

MGVAE/general_graph_generation/create_graphs.py

Line 68 in d254519

for k in range(100):

[HyTruongSon]

Thank you very much for your feedback! I will take a look into the issue when time allows.