feat: Add AGNNConv support for HeteroGraphConv

GNNLux/src/layers/conv.jl

@@ -421,7 +421,7 @@ function Base.show(io::IO, l::AGNNConv)
     print(io, ")")
 end
 
-function (l::AGNNConv)(g, x::AbstractMatrix, ps, st)
+function (l::AGNNConv)(g, x, ps, st)
     β = l.trainable ? ps.β : l.init_beta
     m = (;  β, l.add_self_loops)
     return GNNlib.agnn_conv(m, g, x), st

GNNlib/src/layers/conv.jl

@@ -334,23 +334,34 @@ end
 
 ####################### AGNNConv ######################################
 
-function agnn_conv(l, g::GNNGraph, x::AbstractMatrix)
+function agnn_conv(l, g::AbstractGNNGraph, x)
     check_num_nodes(g, x)
     if l.add_self_loops
         g = add_self_loops(g)
     end
 
-    xn = x ./ sqrt.(sum(x .^ 2, dims = 1))
-    cos_dist = apply_edges(xi_dot_xj, g, xi = xn, xj = xn)
+    xj, xi = expand_srcdst(g, x)
+
+    xi_n = xi ./ sqrt.(sum(xi .^ 2, dims = 1))
+    xj_n = xj ./ sqrt.(sum(xj .^ 2, dims = 1))
+    cos_dist = apply_edges(xi_dot_xj, g, xi = xi_n, xj = xj_n)
     α = softmax_edge_neighbors(g, l.β .* cos_dist)
 
-    x = propagate(g, +; xj = x, e = α) do xi, xj, α
-        α .* xj 
+    x = propagate(g, +; xj, e = α) do xi, xj, α
+        α .* xj
     end
 
     return x
 end
 
+"""
+    _has_same_node_types(g::GNNHeteroGraph)
+
+Return true if all edge types in the heterogeneous graph have the same source and
+target node types (i.e., no bipartite relations).
+"""
+_has_same_node_types(g::GNNHeteroGraph) = all(et -> et[1] == et[3], g.etypes)
+
 ####################### MegNetConv ######################################
 
 function megnet_conv(l, g::GNNGraph, x::AbstractMatrix, e::AbstractMatrix)

GraphNeuralNetworks/test/layers/conv.jl

@@ -422,7 +422,14 @@ end
         g.graph isa AbstractSparseMatrix && continue
         @test size(l(g, g.x)) == (D_IN, g.num_nodes)
         test_gradients(l, g, g.x, rtol = RTOL_HIGH, test_gpu = true, compare_finite_diff = false)
-    end   
+    end
+    l_bip = AGNNConv(add_self_loops=false)
+    s = [1, 1, 2, 3]
+    t = [1, 2, 1, 2]
+    g = GNNGraph((s, t)) |> gpu
+    x = (randn(Float32, D_IN, 3) |> gpu, randn(Float32, D_IN, 2) |> gpu)
+    y = l_bip(g, x)
+    @test size(y) == (D_IN, 2)
 end
 
 @testitem "MEGNetConv" setup=[TolSnippet, TestModule] begin

GraphNeuralNetworks/test/layers/heteroconv.jl

@@ -155,7 +155,15 @@
         x = (A = rand(Float32, 4,2), B = rand(Float32, 4, 3))
         layers = HeteroGraphConv( (:A, :to, :B) => GCNConv(4 => 2, tanh),
                                     (:B, :to, :A) => GCNConv(4 => 2, tanh));
-        y = layers(g, x); 
+        y = layers(g, x);
         @test size(y.A) == (2,2) && size(y.B) == (2,3)
     end
+
+    @testset "AGNNConv" begin
+        x = (A = rand(Float32, 4, 2), B = rand(Float32, 4, 3))
+        layers = HeteroGraphConv((:A, :to, :B) => AGNNConv(),
+                                 (:B, :to, :A) => AGNNConv())
+        y = layers(hg, x)
+        @test size(y.A) == (4, 2) && size(y.B) == (4, 3)
+    end
 end