Batch training on large heterogeneous graph

[zzyzeyuan]

Hi,
I’ m following this dgl docs where has a function named NodeDataLoader(), but actually in my dgl version 1.0.0, there is no such function.

I only have DataLoader, DistDataLoader, DistNodeDataLoader. I wonder which one is corresponding to NodeDataLoader in the docs?

I think DataLoader is what I want, but I am not sure.

And another question is:
If I have a feature lists which contain the features of several type nodes, and the feature dimensions of each node type are different, does it match the batch training in dgl?

[BarclayII]

DataLoader takes the place of NodeDataLoader in 1.0.0.

@zzyzeyuan

If I have a feature lists which contain the features of several type nodes, and the feature dimensions of each node type are different, does it match the batch training in dgl?

I assume you are asking how to train a heterogeneous GNN on a graph whose node types have different kinds of features. A general approach is to just fuse those features into a hidden vector of the same size before applying a heterogeneous GNN (e.g. RGCN).