Run supervised LSTM-based model with default hyperparameters yield an F1 of around 0.46 about graphsage HOT 2 OPEN

Hi,
Have you downloaded the complete data from our website? See http://snap.stanford.edu/graphsage/ for the PPI data.

With adam, the learning rate should be around 0.0005

Also, you can generate the walks file with the deepwalk implementation.

It should converge in ~10 epochs.

from graphsage.

Hi,
Have you downloaded the complete data from our website? See http://snap.stanford.edu/graphsage/ for the PPI data.

With adam, the learning rate should be around 0.0005

Also, you can generate the walks file with the deepwalk implementation.

It should converge in ~10 epochs.

Hi,

I don't really get your point here, since he said he was trying on supervised version, and walks file should only be used in unsupervised version I think.

Also, after tuning the learning rate here to 0.0005(is this really for supervised version again?), my result is not so good.

from graphsage.