Merge pull request #1965 from nouiz/tile

theano/tensor/basic.py

@@ -3999,6 +3999,9 @@ class Tile(Op):
     def __hash__(self):
         return hash(Tile) ^ hash(self.ndim)
 
+    def __str__(self):
+        return self.__class__.__name__ + "{ndim=%d}" % self.ndim
+
     def make_node(self, x, reps):
         x = as_tensor_variable(x)
         reps = as_tensor_variable(reps)
@@ -4054,7 +4057,7 @@ def tile(x, reps, ndim=None):
         assert python_all([int(i) == i for i in iter(reps)])
     except (TypeError, AssertionError):
         raise ValueError("reps argument to tile must be a constant (e.g. "
-        "tuple, list of integers)")
+                         "tuple, list of integers)")
     if len(reps) != x.ndim:
         raise ValueError("len(reps) != x.ndim not currently supported")
     elif (ndim is not None) and ndim != x.ndim:

theano/tensor/opt.py

@@ -2463,6 +2463,44 @@ def local_div_switch_sink(node):
     return False
 
 
+#############
+# Tile Opts #
+#############
+@register_canonicalize
+@register_stabilize
+@gof.local_optimizer([T.Tile])
+def local_useless_tile(node):
+    """Tile(x, (1,)*N) -> x
+
+    This is useless tile. (1,)*N, just mean a vector with all element
+    being 1.
+
+    """
+    if isinstance(node.op, T.Tile):
+        try:
+            a = T.get_scalar_constant_value(node.inputs[1])
+            if a == 1:
+                try:
+                    l = T.get_vector_length(node.inputs[1])
+                    if l == node.inputs[0].ndim:
+                        return [node.inputs[0]]
+                    elif l < node.inputs[0].ndim:
+                        # The Op don't support that case, so we can't
+                        # implement the opt and test it.
+                        return
+                        return [node.inputs[0]]
+                    else:
+                        # The Op don't support that case, so we can't
+                        # implement the opt and test it.
+                        return
+                        x_nd = node.inputs[0].ndim
+                        broad = ['x'] * (l - x_nd) + range(x_nd)
+                        return [node.inputs[0].dimshuffle(broad)]
+                except ValueError:
+                    return
+        except NotScalarConstantError:
+            return
+
 ################
 # Flatten Opts #
 ################

theano/tensor/tests/test_opt.py

@@ -2883,6 +2883,36 @@ def test_local_mul_specialize():
     assert nodes == [T.mul]
 
 
+class T_Tile(unittest.TestCase):
+    def test_local_useless_tile(self):
+        v = T.vector()
+        m = T.matrix()
+        for var, data in [(v, [1, 2, 3]), (m, [[1, 2], [3, 4]])]:
+            # Currently, only a repeat patter == ndim is supported.
+            for ndim in [var.ndim]:  # range(1, var.ndim):
+                f = theano.function([var], T.tile(var, (1,)*ndim))
+                topo = f.maker.fgraph.toposort()
+                assert len(topo) == 1
+                assert isinstance(topo[0].op, compile.DeepCopyOp)
+
+        # If the repeat parameter is longer then v.ndim, we must
+        # replace it with a DimShuffle to add the extra parameter.
+        # But it isn't supported for now, so assert that we raise an
+        # error.
+        self.assertRaises(ValueError, T.tile, v, (1,)*(v.ndim+1))
+        # If the repeat parameter is shorter then m.ndim, it should
+        # pad tot he left the repeat patter with 1. It is not supported for now.
+        #f = theano.function([var], T.tile(v, (1,)*(v.ndim+1)))
+        #topo = f.maker.fgraph.toposort()
+        #assert len(topo) == 1
+        #assert isinstance(topo[0].op, DimShuffe)
+        self.assertRaises(ValueError, T.tile, m, (1,)*(m.ndim-1))
+        #f = theano.function([var], T.tile(m, (1,)*(m.ndim-1)))
+        #topo = f.maker.fgraph.toposort()
+        #assert len(topo) == 1
+        #assert isinstance(topo[0].op, compile.DeepCopyOp)
+
+
 def speed_local_pow_specialize_range():
     val = numpy.random.rand(1e7)
     v = T.vector()