Question about temporal graph neural networks

[DoktorMike]

I was wondering if there is a wish to also include temporal graph neural networks architectures in this repo or if that is preferred to be as a separate package?

There's PyTorch Geometric and PyTorch Geometric Temporal which are currently separate packages if I remember correctly but I wonder if it wouldn't be more natural to have these in the same repo in GraphNeuralNetworks.jl?

Thoughts?

[CarloLucibello]

I would be totally fine with having temporal gnn here. I'm not too worried about possible additional dependencies involved (DataFrames.jl maybe?). Seems a large undertaking though, we have to devise a few new graph types.

[DoktorMike]

I agree that there would not be many more dependencies (if any), but the architectures would be extended by temporal versions of course.

This is what Pytorch Geometric Temporal currently provides I think. https://pytorch-geometric-temporal.readthedocs.io/en/latest/notes/resources.html

Model	Temporal Layer	GNN Layer	Proximity Order	Multi Type
DCRNN	GRU	DiffConv	Higher	False
GConvGRU	GRU	Chebyshev	Lower	False
GConvLSTM	LSTM	Chebyshev	Lower	False
GC-LSTM	LSTM	Chebyshev	Lower	True
DyGrAE	LSTM	GGCN	Higher	False
LRGCN	LSTM	RGCN	Lower	False
EGCN-H	GRU	GCN	Lower	False
EGCN-O	LSTM	GCN	Lower	False
T-GCN	GRU	GCN	Lower	False
A3T-GCN	GRU	GCN	Lower	False
AGCRN	GRU	Chebyshev	Higher	False
MPNN LSTM	LSTM	GCN	Lower	False
STGCN	Attention	Chebyshev	Higher	False
ASTGCN	Attention	Chebyshev	Higher	False
MSTGCN	Attention	Chebyshev	Higher	False
GMAN	Attention	Custom	Lower	False
MTGNN	Attention	Custom	Higher	False
AAGCN	Attention	Custom	Higher	False

[CarloLucibello]

progress in under way, a first graph structure has been implemented in #293

[CarloLucibello]

we now have good support for temporal gnn convs thanks to @aurorarossi