Tengfei Ma

The default parameter did not do any sampling: main(None). Change the "None" into 100 or 200

Thanks. Your update has been merged.

GCN_APPRO: approximate two layers GCN_APPRO_MIX: precomputation in the bottom layer GCN_APPRO_MIX is generally faster and more accurate.

Neither squaring nor nonsquaring guarantees variance reduction, as implied in the paper (see the paragraph above proposition 4). It’s up to empirical tuning to adopt the best sampling distribution. We...

Thanks for transferring it to pytorch. As to the minibatch, I suggest developing only one minibatch version for GCN, so that if you do not want to use minibatch, i.e....

Probably due to the data difference. There is no "val" feature in your data. I believe you can solve it by editing some part of the codes. Thanks.

Thank you for your comment. I have updated the readme file and added two missing files.

the "q" is different. sampleBatch selected elements from all columns while sampleA selected elements only from nonzero columns.

You need to use the code "create_Graph_forGraphSAGE.py" to transfer the format. As per previous comments, please notice your networkx version: nextworkx

It is a two-layer version. But for the bottom layer, we pre-compute the $AX$ as we explained in the paper, so we just did sampling for the top layer.