search

williamleif / GraphSAGE

Representation learning on large graphs using stochastic graph convolutions.
Other
3.43k stars 844 forks source link

why to pad features with zero vector？ #168

Open zhanking321 opened 3 years ago

zhanking321 commented 3 years ago

Excuse me, I wonder why to pad features with zero vector？

if not features is None:
    # pad with dummy zero vector
    features = np.vstack([features, np.zeros((features.shape[1],))])