Weights not included in GNNGraph made from SimpleWeightedDiGraph

[umbriquse]

Hello, just found your sweet package! I ran into a minor issue when wrapping a SimpleWeightedDiGraph with a GNNGraph. The weights from the DiGraph are not included in the wrapped graph.

Below is an MWE of this behavior. I'm not certain if this was by design.

using Graphs, GraphNeuralNetworks, SimpleWeightedGraphs
function randGraph(graphSize::Int)
       graph = rand(graphSize, graphSize)
       foreach(enumerate(eachcol(graph))) do (idx, col)
             graph[idx, :] .= col
             graph[idx, idx] = 0
       end
       return graph
end
a = randGraph(10)
g = SimpleWeightedDiGraph(a)
b = GNNGraph(g)
b.graph

RETURNS

([1, 1, 1, 1, 1, 1, 1, 1, 1, 2  …  9, 10, 10, 10, 10, 10, 10, 10, 10, 10], [2, 3, 4, 5, 6, 7, 8, 9, 10, 1  …  10, 1, 2, 3, 4, 5, 6, 7, 8, 9], nothing)

[CarloLucibello]

SimpleWeightedGraphs.jl is not supported yet. We can add support if think people it is convenient, although I would avoid adding another dependence to this package if possible.

Right now you can create a weighted graph explicitly passing source, target and weights to the constructor:

julia> g = rand_graph(5, 10, bidirected=true)
GNNGraph:
    num_nodes = 5
    num_edges = 10

julia> s, t = edge_index(g)
([1, 1, 2, 2, 4, 4, 5, 4, 5, 5], [4, 5, 4, 5, 5, 1, 1, 2, 2, 4])

julia> w = rand(5);

julia> w = [w; w] # same weight in both directions;

julia> gweighted = GNNGraph(s, t, w)
GNNGraph:
    num_nodes = 5
    num_edges = 10

julia> gweighted.graph
([1, 1, 2, 2, 4, 4, 5, 4, 5, 5], [4, 5, 4, 5, 5, 1, 1, 2, 2, 4], [0.7349537135631546, 0.33820014607400384, 0.08459031049266375, 0.03523075347974425, 0.8861075693177956, 0.7349537135631546, 0.33820014607400384, 0.08459031049266375, 0.03523075347974425, 0.8861075693177956])

Is this convenient enough for you? Also, right now only GCNConv supports edge weight, is that how you were planning to use the weighted graph?

[umbriquse]

Ah, it's not critical for SimpleWeightedGraphs.jl to be supported in my case, just making sure that this was intentional. GCNConv is how I was planning on using edge weights.

[CarloLucibello]

Reopening this as we don't support construction from WeightedGraphs yet

[CarloLucibello]

This could be implemented as a package extension