[Feature] Edge softmax on a subset of edges in the graph. (dmlc#842)

* upd

* add test

* fix

* upd

* merge

* hotfix

* upd

* fix

examples/mxnet/gat/utils.py

@@ -1,5 +1,4 @@
 import numpy as np
-import torch
 
 class EarlyStopping:
     def __init__(self, patience=10):

python/dgl/nn/mxnet/softmax.py

@@ -3,6 +3,7 @@
 import mxnet as mx
 
 from ... import function as fn
+from ...base import ALL, is_all
 
 __all__ = ['edge_softmax']
 
@@ -24,8 +25,10 @@ class EdgeSoftmax(mx.autograd.Function):
     the attention weights are computed with such an edgesoftmax operation.
     """
 
-    def __init__(self, g):
+    def __init__(self, g, eids):
         super(EdgeSoftmax, self).__init__()
+        if not is_all(eids):
+            g = g.edge_subgraph(eids.astype('int64'))
         self.g = g
 
     def forward(self, score):
@@ -78,7 +81,7 @@ def backward(self, grad_out):
         grad_score = g.edata['grad_score'] - g.edata['out']
         return grad_score
 
-def edge_softmax(graph, logits):
+def edge_softmax(graph, logits, eids=ALL):
     r"""Compute edge softmax.
 
     For a node :math:`i`, edge softmax is an operation of computing
@@ -98,8 +101,11 @@ def edge_softmax(graph, logits):
     ----------
     graph : DGLGraph
         The graph to perform edge softmax
-    logits : torch.Tensor
+    logits : mxnet.NDArray
         The input edge feature
+    eids : mxnet.NDArray or ALL, optional
+        Edges on which to apply edge softmax. If ALL, apply edge softmax
+        on all edges in the graph. Default: ALL.
 
     Returns
     -------
@@ -108,9 +114,10 @@ def edge_softmax(graph, logits):
 
     Notes
     -----
-        * Input shape: :math:`(N, *, 1)` where * means any number of
-          additional dimensions, :math:`N` is the number of edges.
-        * Return shape: :math:`(N, *, 1)`
+        * Input shape: :math:`(E, *, 1)` where * means any number of
+          additional dimensions, :math:`E` equals the length of eids.
+          If eids is ALL, :math:`E` equals number of edges in the graph.
+        * Return shape: :math:`(E, *, 1)`
 
     Examples
     --------
@@ -143,6 +150,14 @@ def edge_softmax(graph, logits):
      [0.33333334]
      [0.33333334]]
     <NDArray 6x1 @cpu(0)>
+
+    Apply edge softmax on first 4 edges of g:
+    >>> edge_softmax(g, edata, nd.array([0,1,2,3], dtype='int64'))
+    [[1. ]
+     [0.5]
+     [1. ]
+     [0.5]]
+    <NDArray 4x1 @cpu(0)>
     """
-    softmax_op = EdgeSoftmax(graph)
+    softmax_op = EdgeSoftmax(graph, eids)
     return softmax_op(logits)

python/dgl/nn/pytorch/softmax.py

@@ -3,6 +3,7 @@
 import torch as th
 
 from ... import function as fn
+from ...base import ALL, is_all
 
 __all__ = ['edge_softmax']
 
@@ -25,7 +26,7 @@ class EdgeSoftmax(th.autograd.Function):
     """
 
     @staticmethod
-    def forward(ctx, g, score):
+    def forward(ctx, g, score, eids):
         """Forward function.
 
         Pseudo-code:
@@ -41,6 +42,8 @@ def forward(ctx, g, score):
         """
         # remember to save the graph to backward cache before making it
         # a local variable
+        if not is_all(eids):
+            g = g.edge_subgraph(eids.long())
         ctx.backward_cache = g
         g = g.local_var()
         g.edata['s'] = score
@@ -79,10 +82,10 @@ def backward(ctx, grad_out):
         g.update_all(fn.copy_e('grad_s', 'm'), fn.sum('m', 'accum'))
         g.apply_edges(fn.e_mul_v('out', 'accum', 'out'))
         grad_score = g.edata['grad_s'] - g.edata['out']
-        return None, grad_score
+        return None, grad_score, None
 
 
-def edge_softmax(graph, logits):
+def edge_softmax(graph, logits, eids=ALL):
     r"""Compute edge softmax.
 
     For a node :math:`i`, edge softmax is an operation of computing
@@ -104,6 +107,9 @@ def edge_softmax(graph, logits):
         The graph to perform edge softmax
     logits : torch.Tensor
         The input edge feature
+    eids : torch.Tensor or ALL, optional
+        Edges on which to apply edge softmax. If ALL, apply edge
+        softmax on all edges in the graph. Default: ALL.
 
     Returns
     -------
@@ -112,9 +118,10 @@ def edge_softmax(graph, logits):
 
     Notes
     -----
-        * Input shape: :math:`(N, *, 1)` where * means any number of
-          additional dimensions, :math:`N` is the number of edges.
-        * Return shape: :math:`(N, *, 1)`
+        * Input shape: :math:`(E, *, 1)` where * means any number of
+          additional dimensions, :math:`E` equals the length of eids.
+          If eids is ALL, :math:`E` equals number of edges in the graph.
+        * Return shape: :math:`(E, *, 1)`
 
     Examples
     --------
@@ -145,5 +152,12 @@ def edge_softmax(graph, logits):
         [0.5000],
         [0.3333],
         [0.3333]])
+
+    Apply edge softmax on first 4 edges of g:
+    >>> edge_softmax(g, edata[:4], th.Tensor([0,1,2,3]))
+    tensor([[1.0000],
+        [0.5000],
+        [1.0000],
+        [0.5000]])
     """
-    return EdgeSoftmax.apply(graph, logits)
+    return EdgeSoftmax.apply(graph, logits, eids)

tests/mxnet/test_nn.py

@@ -223,6 +223,36 @@ def test_edge_softmax():
     assert np.allclose(a.asnumpy(), uniform_attention(g, a.shape).asnumpy(),
             1e-4, 1e-4)
 
+def test_partial_edge_softmax():
+    g = dgl.DGLGraph()
+    g.add_nodes(30)
+    # build a complete graph
+    for i in range(30):
+        for j in range(30):
+            g.add_edge(i, j)
+
+    score = F.randn((300, 1))
+    score.attach_grad()
+    grad = F.randn((300, 1))
+    import numpy as np
+    eids = np.random.choice(900, 300, replace=False).astype('int64')
+    eids = F.zerocopy_from_numpy(eids)
+    # compute partial edge softmax
+    with mx.autograd.record():
+        y_1 = nn.edge_softmax(g, score, eids)
+        y_1.backward(grad)
+        grad_1 = score.grad
+
+    # compute edge softmax on edge subgraph
+    subg = g.edge_subgraph(eids)
+    with mx.autograd.record():
+        y_2 = nn.edge_softmax(subg, score)
+        y_2.backward(grad)
+        grad_2 = score.grad
+
+    assert F.allclose(y_1, y_2)
+    assert F.allclose(grad_1, grad_2)
+
 def test_rgcn():
     ctx = F.ctx()
     etype = []
@@ -277,6 +307,7 @@ def test_rgcn():
 if __name__ == '__main__':
     test_graph_conv()
     test_edge_softmax()
+    test_partial_edge_softmax()
     test_set2set()
     test_glob_att_pool()
     test_simple_pool()

tests/pytorch/test_nn.py

@@ -319,7 +319,36 @@ def generate_rand_graph(n):
     assert len(g.ndata) == 0
     assert len(g.edata) == 2
     assert F.allclose(a1.grad, a2.grad, rtol=1e-4, atol=1e-4) # Follow tolerance in unittest backend
-
+
+def test_partial_edge_softmax():
+    g = dgl.DGLGraph()
+    g.add_nodes(30)
+    # build a complete graph
+    for i in range(30):
+        for j in range(30):
+            g.add_edge(i, j)
+
+    score = F.randn((300, 1))
+    score.requires_grad_()
+    grad = F.randn((300, 1))
+    import numpy as np
+    eids = np.random.choice(900, 300, replace=False).astype('int64')
+    eids = F.zerocopy_from_numpy(eids)
+    # compute partial edge softmax
+    y_1 = nn.edge_softmax(g, score, eids)
+    y_1.backward(grad)
+    grad_1 = score.grad
+    score.grad.zero_()
+    # compute edge softmax on edge subgraph
+    subg = g.edge_subgraph(eids)
+    y_2 = nn.edge_softmax(subg, score)
+    y_2.backward(grad)
+    grad_2 = score.grad
+    score.grad.zero_()
+
+    assert F.allclose(y_1, y_2)
+    assert F.allclose(grad_1, grad_2)
+
 def test_rgcn():
     ctx = F.ctx()
     etype = []
@@ -570,6 +599,7 @@ def test_dense_cheb_conv():
 if __name__ == '__main__':
     test_graph_conv()
     test_edge_softmax()
+    test_partial_edge_softmax()
     test_set2set()
     test_glob_att_pool()
     test_simple_pool()