danielegrattarola
commented
4 years ago Hi,
GCNs usually have the node features as input and output. What you are trying to achieve sounds like a link prediction task. Usually, this is done by first learning a representation of the node features with a GNN, and then using some similarity function to predict the presence/absence of a link.
A very simple approach to this is the variational graph autoencoder by Kipf et al., where they compute a representation Z and then do the link prediction as sigmoid(Z*Z^T). If you want to predict a continuous variable, you can remove the sigmoid or do something different, but the idea is that your output should be a function of node pairs.
Hope this helps. Best, Daniele
Hi Daniele, I am trying to use Graph Convolutional Networks to forecast traffic flows between origin and destination cities. I only have one graph so it should be a single mode. I am wondering is it possible to get a matrix format output (N*N) using GraphConv layer and LSTM layer? Will this output considering both the interactions between origin cities and destination cities? I will really appreciate your kind help. Thanks!