[Bug] Do not skip graphconv even no edge exists (#3416)

python/dgl/nn/mxnet/hetero.py

@@ -165,8 +165,6 @@ class HeteroGraphConv(nn.Block):
             src_inputs, dst_inputs = inputs
             for stype, etype, dtype in g.canonical_etypes:
                 rel_graph = g[stype, etype, dtype]
-                if rel_graph.number_of_edges() == 0:
-                    continue
                 if stype not in src_inputs or dtype not in dst_inputs:
                     continue
                 dstdata = self.mods[etype](
@@ -178,8 +176,6 @@ class HeteroGraphConv(nn.Block):
         else:
             for stype, etype, dtype in g.canonical_etypes:
                 rel_graph = g[stype, etype, dtype]
-                if rel_graph.number_of_edges() == 0:
-                    continue
                 if stype not in inputs:
                     continue
                 dstdata = self.mods[etype](

python/dgl/nn/pytorch/hetero.py

@@ -169,8 +169,6 @@ class HeteroGraphConv(nn.Module):
 
             for stype, etype, dtype in g.canonical_etypes:
                 rel_graph = g[stype, etype, dtype]
-                if rel_graph.number_of_edges() == 0:
-                    continue
                 if stype not in src_inputs or dtype not in dst_inputs:
                     continue
                 dstdata = self.mods[etype](
@@ -182,8 +180,6 @@ class HeteroGraphConv(nn.Module):
         else:
             for stype, etype, dtype in g.canonical_etypes:
                 rel_graph = g[stype, etype, dtype]
-                if rel_graph.number_of_edges() == 0:
-                    continue
                 if stype not in inputs:
                     continue
                 dstdata = self.mods[etype](

python/dgl/nn/tensorflow/hetero.py

@@ -169,8 +169,6 @@ class HeteroGraphConv(layers.Layer):
             src_inputs, dst_inputs = inputs
             for stype, etype, dtype in g.canonical_etypes:
                 rel_graph = g[stype, etype, dtype]
-                if rel_graph.number_of_edges() == 0:
-                    continue
                 if stype not in src_inputs or dtype not in dst_inputs:
                     continue
                 dstdata = self.mods[etype](
@@ -182,8 +180,6 @@ class HeteroGraphConv(layers.Layer):
         else:
             for stype, etype, dtype in g.canonical_etypes:
                 rel_graph = g[stype, etype, dtype]
-                if rel_graph.number_of_edges() == 0:
-                    continue
                 if stype not in inputs:
                     continue
                 dstdata = self.mods[etype](

tests/mxnet/test_nn.py

@@ -788,6 +788,19 @@ def test_hetero_conv(agg, idtype):
     assert mod2.carg1 == 1
     assert mod3.carg1 == 0
 
+    #conv on graph without any edges
+    for etype in g.etypes:
+        g = dgl.remove_edges(g, g.edges(form='eid', etype=etype), etype=etype)
+    assert g.num_edges() == 0
+    h = conv(g, {'user': uf, 'game': gf, 'store': sf})
+    assert set(h.keys()) == {'user', 'game'}
+
+    block = dgl.to_block(g.to(F.cpu()), {'user': [0, 1, 2, 3], 'game': [
+                         0, 1, 2, 3], 'store': []}).to(F.ctx())
+    h = conv(block, ({'user': uf, 'game': gf, 'store': sf},
+             {'user': uf, 'game': gf, 'store': sf[0:0]}))
+    assert set(h.keys()) == {'user', 'game'}
+
 if __name__ == '__main__':
     test_graph_conv()
     test_gat_conv()
@@ -809,3 +822,4 @@ if __name__ == '__main__':
     test_simple_pool()
     test_rgcn()
     test_sequential()
+    test_hetero_conv()

tests/pytorch/test_nn.py

@@ -1112,6 +1112,19 @@ def test_hetero_conv(agg, idtype):
     assert mod3.carg1 == 0
     assert mod3.carg2 == 1
 
+    #conv on graph without any edges
+    for etype in g.etypes:
+        g = dgl.remove_edges(g, g.edges(form='eid', etype=etype), etype=etype)
+    assert g.num_edges() == 0
+    h = conv(g, {'user': uf, 'game': gf, 'store': sf})
+    assert set(h.keys()) == {'user', 'game'}
+
+    block = dgl.to_block(g.to(F.cpu()), {'user': [0, 1, 2, 3], 'game': [
+                         0, 1, 2, 3], 'store': []}).to(F.ctx())
+    h = conv(block, ({'user': uf, 'game': gf, 'store': sf},
+             {'user': uf, 'game': gf, 'store': sf[0:0]}))
+    assert set(h.keys()) == {'user', 'game'}
+
 if __name__ == '__main__':
     test_graph_conv()
     test_graph_conv_e_weight()
@@ -1140,3 +1153,4 @@ if __name__ == '__main__':
     test_sequential()
     test_atomic_conv()
     test_cf_conv()
+    test_hetero_conv()

tests/tensorflow/test_nn.py

@@ -502,6 +502,19 @@ def test_hetero_conv(agg, idtype):
     assert mod3.carg1 == 0
     assert mod3.carg2 == 1
 
+    #conv on graph without any edges
+    for etype in g.etypes:
+        g = dgl.remove_edges(g, g.edges(form='eid', etype=etype), etype=etype)
+    assert g.num_edges() == 0
+    h = conv(g, {'user': uf, 'game': gf, 'store': sf})
+    assert set(h.keys()) == {'user', 'game'}
+
+    block = dgl.to_block(g.to(F.cpu()), {'user': [0, 1, 2, 3], 'game': [
+                         0, 1, 2, 3], 'store': []}).to(F.ctx())
+    h = conv(block, ({'user': uf, 'game': gf, 'store': sf},
+             {'user': uf, 'game': gf, 'store': sf[0:0]}))
+    assert set(h.keys()) == {'user', 'game'}
+
 
 @pytest.mark.parametrize('out_dim', [1, 2])
 def test_dense_cheb_conv(out_dim):
@@ -549,3 +562,4 @@ if __name__ == '__main__':
     # test_dense_sage_conv()
     test_dense_cheb_conv()
     # test_sequential()
+    test_hetero_conv()