Missing functionality compared to DGL

[CarloLucibello]

Checklist of stuff we miss compared to Deep Graph Library.
PRs are welcome!

Conv Layers

• GraphConv (called GCNConv here)
• EdgeWeightNorm
• RelGraphConv
• TAGConv
• GATConv
• EdgeConv
• SAGEConv
• SGConv
• APPNPConv
• GINConv
• GatedGraphConv
• GMMConv (Added GMMConv #147)
• ChebConv
• AGNNConv
• NNConv
• AtomicConv
• CFConv
• DotGatConv
• TWIRLSConv
• TWIRLSUnfoldingAndAttention
• GCN2Conv

Dense Conv Layers

• DenseGraphConv
• DenseSAGEConv
• DenseChebConv

Global Pooling Layers

• SumPooling (GlobalPooling(+) here)
• AvgPooling (GlobalPooling(mean) here)
• MaxPooling (GlobalPooling(max) here)
• SortPooling
• WeightAndSum
• GlobalAttentionPooling
• Set2Set
• SetTransformerEncoder
• SetTransformerDecoder

Batching and Reading Out Ops

https://docs.dgl.ai/en/0.6.x/api/python/dgl.html#batching-and-reading-out-ops

• batch. Use Flux.batch or SparseArrays.blockdiag
• unbatch
• readout_nodes (called reduce_nodes here)
• readout_edges (called reduce_edges here)
• sum_nodes # use reduce_nodes(+, g, x)
• sum_edges # use reduce_edges(+, g, x)
• mean_nodes
• mean_edges
• max_nodes
• max_edges
• softmax_nodes
• softmax_edges
• broadcast_nodes
• broadcast_edges
• topk_nodes
• topk_edges

Adjacency Related Utilities

• khop_adj
• laplacian_lambda_max

nn.functional

https://docs.dgl.ai/api/python/nn.functional.html

• edge_softmax (softmax_edge_neighbors here)

optim

https://docs.dgl.ai/api/python/dgl.optim.html

• Sparse Adam
• Sparse AdaGrad

nn Utility Modules

• Sequential (GNNChain here)
• WeightBasis
• KNNGraph
• SegmentedKNNGraph

nn NodeEmbedding Module

• NodeEmbedding

Sampling and Stochastic training

.....

Distributed Training

....

[oysteinsolheim]

Would it be possible to also have a list of wished-for-algorithms, included or not in DGL? :-) For example I'd love to see edge-featured-based algorithms in general, and maybe more specifically the "Exploiting Edge Features in Graph Neural Networks" from .https://arxiv.org/abs/1809.02709

[CarloLucibello]

Of course! You can file a separate issue for each feature request and I'll try to get to them as soon as I can spare some time if no one beats me to it. For instance, "Exploiting Edge Features in Graph Neural Networks" seems to be cited enough that it is worth having, so you are welcome to open a new issue.

[oysteinsolheim]

Perfect! :-)