GCNSVD predict() about deeprobust HOT 4 CLOSED

@jiong-zhu Yeah, that makes sense. I've updated the code as you suggested to avoid confusion. See details in commit 000f86d. Thank you for the advice, Jiong and Mark!

from deeprobust.

Thanks for the suggestion.

Typically self.predict() is used by self.test() where we do not specify the arguments features and adj. If you hope to pass some features, adj pair into the function, you can do it outside the predict() as follows,

prediction1 = model.predict()
processed_adj = model.truncatedSVD(perturbed_adj, k=20)
# prediction2 is the same as prediction1
prediction2 = model.predict(features, processed_adj)

Similarly we can do that in GCNJaccard. I think I am going to keep the original version of predict(). Thank you again for your suggestions and feel free to let me know if you have other concerns.

from deeprobust.

@ChandlerBang I see your point, but I would still vote to include an overridden version of the predict function for GCNSVD, just to keep the consistency of interfaces among different models. If kept unchanged, users using the predict() function for GCNSVD may think they are getting predictions on the GCNSVD, but in fact they are getting results from GCN.

from deeprobust.

Thank you!

from deeprobust.