search

plai-group / gae_in_pytorch

Graph Auto-Encoder in PyTorch
Apache License 2.0
81 stars 18 forks source link

Ismember function #7

Open haorannlp opened 4 years ago

haorannlp commented 4 years ago

Hi！ I notice that in your implementation, the "Ismember" function in preprocessing.py is different from Tkipf's. I tested a few toy examples, found that your implementation would wrongly classify a member edge as a non-member edge. Since the whole adjacency matrix is sparse, the overall performance is not affected by this bug. Could you pls help me check it out? Thank you so much.

Dzhilin commented 3 years ago

Hello, I would like to ask you what's the use of "tol = 5" in "ismember" function and "- 1" in np.all(..., axis = - 1)? Looking forward to your answer:)

vmasrani commented 3 years ago

tol = 5 rounds to the fifth decimal place in np.round and np.all(..., axis = - 1) checks to see if all the entries along the last axis evaluate to True. Taken together np.all(np.round(a - b[:, None], tol) == 0, axis=-1) will be True if all entries of a and b[:, None] are the same (within five decimal of precision) and False otherwise