added ShuffleRule

elemwise.py

@@ -2,7 +2,7 @@
 import elemwise_cgen as cgen
 
 import numpy
-from gof import Op, Apply
+from gof import Op, Macro, Apply
 import scalar
 from scalar import Scalar
 import gof
@@ -29,6 +29,50 @@ def TensorConstant(*inputs, **kwargs):
 ### DimShuffle ###
 ##################
 
+## TODO: rule-based version of DimShuffle
+## would allow for Transpose, LComplete, RComplete, etc.
+## Can be optimized into DimShuffle later on.
+
+
+class ShuffleRule(Macro):
+    """
+    ABSTRACT Op - it has no perform and no c_code
+    
+    Apply ExpandMacros to this node to obtain
+    an equivalent DimShuffle which can be performed.
+    """
+    def __init__(self, rule = None, name = None):
+        if rule is not None:
+            self.rule = rule
+        self.name = name
+    def make_node(self, input, *models):
+        pattern = self.rule(input.type.broadcastable, *(model.type.broadcastable for model in models))
+        return gof.Apply(self,
+                         (input,) + models,
+                         [Tensor(dtype = input.type.dtype,
+                                 broadcastable = [x == 'x' for x in pattern]).make_result()])
+    def expand(self, r):
+        input, models = r.owner.inputs[0], r.owner.inputs[1:]
+        new_order = self.rule(input.type.broadcastable, *(model.type.broadcastable for model in models))
+        return DimShuffle(input.type.broadcastable, new_order)(input)
+    def __eq__(self, other):
+        return type(self) == type(other) and self.rule == other.rule
+    def __hash__(self, other):
+        return hash(self.rule)
+    def __str__(self):
+        if self.name is not None:
+            return self.name
+        else:
+            return "ShuffleRule{%s}" % self.role
+
+_transpose = ShuffleRule(rule = lambda input: range(len(input)-1, -1, -1),
+                         name = 'transpose')
+lcomplete = ShuffleRule(rule = lambda input, *models: ['x']*(max([0]+map(len,models))-len(input)) + range(len(input)),
+                        name = 'lcomplete')
+rcomplete = ShuffleRule(rule = lambda input, *models: range(len(input)) + ['x']*(max(map(len,models))-len(input)),
+                        name = 'rcomplete')
+
+
 
 class DimShuffle(Op):
     """
@@ -182,6 +226,23 @@ class DimShuffle(Op):
         return DimShuffle(gz.type.broadcastable, grad_order)(gz),
 
 
+# class LComplete(Op):
+#     view_map = {0: [0]}
+#     def make_node(self, x, y):
+#         x, y = map(as_tensor, (x, y))
+#         xd, yd = x.type.ndim, y.type.ndim
+#         if xd > yd:
+#             raise TypeError("The tensor to left-complete has more dimensions than the model.")
+#         return gof.Apply(self,
+#                          [x, y],
+#                          [Tensor(dtype = x.type.dtype,
+#                                  broadcastable = (True,)*(yd-xd) + x.type.broadcastable).make_result()])
+#     def perform(self, node, (x, y), (z, )):
+#         return x.reshape((1, )*(y.ndim - x.ndim) + tuple(x.shape))
+#     def grad(self, node, (x, ), (gz, )):
+#         xd, gzd = x.type.ndim, gz.type.ndim
+#         return DimShuffle(gz.broadcastable, range(gzd-xd, xd))(gz)
+
 
 ################
 ### Elemwise ###
@@ -243,20 +304,23 @@ class Elemwise(Op):
         shadow = self.scalar_op.make_node(*[Scalar(dtype = t.type.dtype)() for t in inputs])
 
         target_length = max([input.type.ndim for input in inputs])
-        args = []
-        for input in inputs:
-            length = input.type.ndim
-            difference = target_length - length
-            if not difference:
-                args.append(input)
-            else:
-                args.append(DimShuffle(range(length), ['x']*difference + range(length))(input))
-        inputs = args
-
-#         try:
-#             assert len(set([len(input.type.broadcastable) for input in inputs])) == 1
-#         except (AssertionError, AttributeError):
-#             raise TypeError("All inputs to a Broadcast subclass must be Tensor instances and their broadcastable fields must all have the same length.", inputs)
+        if len(inputs) > 1:
+            inputs = [lcomplete(input, *inputs) for input in inputs]
+#         args = []
+#         for input in inputs:
+#             length = input.type.ndim
+#             difference = target_length - length
+#             if not difference:
+#                 args.append(input)
+#             else:
+#                 # TODO: use LComplete instead
+#                 args.append(DimShuffle(input.type.broadcastable, ['x']*difference + range(length))(input))
+#         inputs = args
+
+        try:
+            assert len(set([len(input.type.broadcastable) for input in inputs])) == 1
+        except (AssertionError, AttributeError):
+            raise TypeError("All inputs to a Broadcast subclass must be Tensor instances and their broadcastable fields must all have the same length.", inputs)
 
         out_broadcastables = [[all(bcast) for bcast in zip(*[input.type.broadcastable for input in inputs])]] * shadow.nout
         inplace_pattern = self.inplace_pattern
@@ -509,6 +573,12 @@ class CAReduce(Op):
                         broadcastable = [x for i, x in enumerate(input.type.broadcastable) if i not in axis])()
         return Apply(self, [input], [output])
 
+    def __eq__(self, other):
+        return type(self) == type(other) and self.scalar_op == other.scalar_op and self.axis == other.axis
+
+    def __hash__(self):
+        return hash(self.scalar_op) ^ hash(self.axis)
+    
     def __str__(self):
         if self.axis is not None:
             return "Reduce{%s}{%s}" % (self.scalar_op, ", ".join(str(x) for x in self.axis))