Merge pull request #445 from lamblin/fix_test_structureddot_grad

theano/sparse/basic.py

@@ -1666,17 +1666,10 @@ def structured_dot_grad(sparse_A, dense_B, ga):
 
         if sparse_A.type.format == 'csc':
             sdgcsx = sdg_csc
-        else:
-            sdgcsx = sdg_csr
-        #backport
-        #sdgcsx = sdg_csc if sparse_A.type.format == 'csc' else sdg_csr
-
-        if sparse_A.type.format == 'csc':
             CSx = CSC
         else:
+            sdgcsx = sdg_csr
             CSx = CSR
-        #backport
-        #CSx = CSC if sparse_A.type.format == 'csc' else CSR
 
         g_A_data = sdgcsx(csm_indices(sparse_A), \
                           csm_indptr(sparse_A), dense_B, ga)
@@ -1705,7 +1698,13 @@ class StructuredDotGradCSC(gof.Op):
             ind1 = a_indptr[j + 1]
             for i_idx in xrange(ind0, ind1):
                 i = a_indices[i_idx]
-                g_a_data[i_idx] = numpy.dot(g_ab[i], b[j].T)[0, 0]
+                # Depending on the type of g_ab and b (sparse or dense),
+                # the following dot product can result in a scalar or
+                # a (1, 1) sparse matrix.
+                dot_val = numpy.dot(g_ab[i], b[j].T)
+                if isinstance(dot_val, scipy.sparse.spmatrix):
+                    dot_val = dot_val[0, 0]
+                g_a_data[i_idx] = dot_val
         out[0] = g_a_data
 
     def c_code(self, node, name, (_indices, _indptr, _d, _g), (_zout, ), sub):
@@ -1820,7 +1819,13 @@ class StructuredDotGradCSR(gof.Op):
             for j_idx in xrange(ind0, ind1):
                 j = a_indices[j_idx]
                 # grad is dot product of i-th row of gradient with j-th row of b
-                g_a_data[j_idx] = numpy.dot(g_ab[i], b[j].T)[0, 0]
+                # Depending on the type of g_ab and b (sparse or dense),
+                # the following dot product can result in a scalar or
+                # a (1, 1) sparse matrix.
+                dot_val = numpy.dot(g_ab[i], b[j].T)
+                if isinstance(dot_val, scipy.sparse.spmatrix):
+                    dot_val = dot_val[0, 0]
+                g_a_data[j_idx] = dot_val
         out[0] = g_a_data
 
     def c_code(self, node, name, (_indices, _indptr, _d, _g), (_zout, ), sub):

theano/sparse/tests/test_basic.py

@@ -448,6 +448,17 @@ class test_structureddot(unittest.TestCase):
         utt.verify_grad(buildgraphCSC,
                     [spmat.data, mat])
 
+        def buildgraphCSC_T(spdata, sym_mat):
+            csc = CSC(spdata, spmat.indices[:spmat.size],
+                    spmat.indptr, spmat.shape)
+            assert csc.type.dtype == 'float32'
+            rval = structured_dot(sym_mat.T, csc.T)
+            assert rval.type.dtype == 'float32'
+            return rval
+
+        utt.verify_grad(buildgraphCSC_T,
+                    [spmat.data, mat])
+
     def test_structureddot_csr_grad(self):
 
         #shortcut: testing csc in float32, testing csr in float64
@@ -468,6 +479,17 @@ class test_structureddot(unittest.TestCase):
         utt.verify_grad(buildgraph,
                     [spmat.data, mat])
 
+        def buildgraph_T(spdata, sym_mat):
+            csr = CSR(spdata, spmat.indices[:spmat.size],
+                    spmat.indptr, spmat.shape)
+            assert csr.type.dtype == 'float64'
+            rval = structured_dot(sym_mat.T, csr.T)
+            assert rval.type.dtype == 'float64'
+            return rval
+
+        utt.verify_grad(buildgraph,
+                    [spmat.data, mat])
+
     def test_infer_shape_csr_csc_grad(self):
         for sparsetype in ('csr', 'csc'):
             a = SparseType(sparsetype, dtype=config.floatX)()