Add theano.sparse_grad interface.

doc/library/index.txt

@@ -48,3 +48,5 @@ There are also some top-level imports that you might find more convenient:
     Works like :func:`tensor.dot` for both sparse and dense matrix products
 
 .. autofunction:: theano.clone
+
+.. autofunction:: theano.sparse_grad

theano/__init__.py

@@ -168,6 +168,20 @@ def get_scalar_constant_value(v):
     return tensor.get_scalar_constant_value(v)
 
 
+def sparse_grad(var):
+    """This function return a new variable whose gradient will be
+    stored in a sparse format instead of dense.
+
+    Currently only AdvancedSubtensor1 is supported.
+
+    """
+    assert isinstance(var.owner.op, tensor.AdvancedSubtensor1)
+    # TODO change the internal representation!!!
+    # It work, but bad as out.type is shared with var.type!!!
+    var.owner.inputs[0].tag.sparse_grad = True
+    return var
+
+
 import theano.tests
 if hasattr(theano.tests, "TheanoNoseTester"):
     test = theano.tests.TheanoNoseTester().test

theano/sparse/tests/test_basic.py

@@ -418,28 +418,46 @@ class SparseInferShapeTester(utt.InferShapeTester):
 
 class TestConstructSparseFromList(unittest.TestCase):
     def test_adv_sub1_sparse_grad(self):
-        m = theano.tensor.matrix()
         v = theano.tensor.ivector()
-        sub = m[v]
 
         # Assert we don't create a sparse grad by default
+        m = theano.tensor.matrix()
+        sub = m[v]
         g = theano.grad(sub.sum(), m)
         assert isinstance(g.owner.op, tensor.AdvancedIncSubtensor1)
 
         # Test that we create a sparse grad when asked
+        # OLD INTERFACE
+        m = theano.tensor.matrix()
+        sub = m[v]
+        m.type.sparse_grad = True
+        g = theano.grad(sub.sum(), m)
+        assert isinstance(g.owner.op, ConstructSparseFromList)
+
+        # Test that we create a sparse grad when asked
+        # OLD INTERFACE CONSEQUENCE
+        m = theano.tensor.matrix()
+        sub = m[v]
         sub.type.sparse_grad = True
         g = theano.grad(sub.sum(), m)
         assert isinstance(g.owner.op, ConstructSparseFromList)
 
+        # Test that we create a sparse grad when asked
+        # USER INTERFACE
+        m = theano.tensor.matrix()
+        v = theano.tensor.ivector()
+        sub = theano.sparse_grad(m[v])
+        g = theano.grad(sub.sum(), m)
+        assert isinstance(g.owner.op, ConstructSparseFromList)
+
         # Test the sparse grad
         valm = numpy.random.rand(5, 4).astype(config.floatX)
         valv = numpy.random.random_integers(0, 4, 10)
+        m = theano.tensor.matrix()
         shared_v = theano.shared(valv)
 
         def fn(m):
-            sub = m[shared_v]
-            sub.type.sparse_grad = True
-            return sub
+            return theano.sparse_grad(m[shared_v])
         verify_grad_sparse(fn, [valm])
 
     def test_err(self):
@@ -454,7 +472,7 @@ class TestConstructSparseFromList(unittest.TestCase):
             assert isinstance(g.owner.op, tensor.AdvancedIncSubtensor1)
 
             # Test that we create a sparse grad when asked
-            sub.type.sparse_grad = True
+            sub = theano.sparse_grad(sub)
             self.assertRaises(TypeError, theano.grad, sub.sum(), t)