GCN process Batched Graph

I tried to run a simple GraphConv on my graph dataloader. Everytime it is batched and I wnat to print out the output feature of GraphConv it leads to a DGLError: There are 0-in-degree nodes in the graph, output for those nodes will be invalid. The error is not there when I pass in a single Graph. What might be the cause? Every graph would have different number of nodes and edges in my case.

Could you check the in_degree numbers of the batched graph and the single graphs in that batch? I think it may no be an issue of batching operation. For graphs that have 0-in-degree nodes, you can add self loops to avoid this error.