Setting of batch size and sampled layer size for large network

[LynnZhang9]

Thank you for your nice work.
Now I try to replace GCN model with Fast GCN model because the graph is a large network (around 2973489 nodes with '0,1' two labels). There are only around 2000 nodes with label '1'.

When I use default setting for such a large graph, it's hard to learn something for Fast GCN model.

For those strong imbalance graph network, how should I set the parameter, like layers, dimension of hidden layer, batch size and sampled size for each layer?
I would be appreciated if you give some hints or advice!