Thank you for making the code available. I would like to ask a question about a remark you made in your paper about the PPI dataset. On page 8, paragraph "Inductive learning on multiple graphs", you noted that 40% of the node has all-zero feature vector. However, the feature vectors loaded using GraphSAGE (http://snap.stanford.edu/graphsage/) is dense. Did you use a different set of feature vectors or a different PPI dataset?
Thank you for your time! If I misunderstood something, please kindly point out my mistake.
Edit: Sorry I made a mistake when I check the feature vectors. It was indeed 42% all zeros.
Hi,
Thank you for making the code available. I would like to ask a question about a remark you made in your paper about the PPI dataset. On page 8, paragraph "Inductive learning on multiple graphs", you noted that 40% of the node has all-zero feature vector. However, the feature vectors loaded using GraphSAGE (http://snap.stanford.edu/graphsage/) is dense. Did you use a different set of feature vectors or a different PPI dataset?
Thank you for your time! If I misunderstood something, please kindly point out my mistake.
Edit: Sorry I made a mistake when I check the feature vectors. It was indeed 42% all zeros.