Merge pull request #8843 from zhouhanqing/Paddle-ReduceProd

Add product reduction for reduce op.

Author: lcy-seso

paddle/fluid/operators/reduce_op.cc

@@ -173,6 +173,15 @@ class ReduceMinOpMaker : public ReduceOpMaker {
   }
 };
 
+class ReduceProdOpMaker : public ReduceOpMaker {
+ public:
+  ReduceProdOpMaker(OpProto *proto, OpAttrChecker *op_checker)
+      : ReduceOpMaker(proto, op_checker) {
+    SetComment("ReduceProd", "production");
+    AddComment(comment_);
+  }
+};
+
 }  // namespace operators
 }  // namespace paddle
 
@@ -190,6 +199,9 @@ REGISTER_OP(reduce_max, ops::ReduceOp, ops::ReduceMaxOpMaker, reduce_max_grad,
 REGISTER_OP(reduce_min, ops::ReduceOp, ops::ReduceMinOpMaker, reduce_min_grad,
             ops::ReduceGradOp);
 
+REGISTER_OP(reduce_prod, ops::ReduceOp, ops::ReduceProdOpMaker,
+            reduce_prod_grad, ops::ReduceGradOp);
+
 #define REGISTER_REDUCE_CPU_KERNEL(reduce_type, functor, grad_functor)         \
   REGISTER_OP_CPU_KERNEL(reduce_type,                                          \
                          ops::ReduceKernel<paddle::platform::CPUDeviceContext, \

paddle/fluid/operators/reduce_op.h

@@ -93,6 +93,22 @@ struct MaxOrMinGradFunctor {
   }
 };
 
+struct ProdFunctor {
+  template <typename DeviceContext, typename X, typename Y, typename Dim>
+  void operator()(const DeviceContext& place, X& x, Y& y, const Dim& dim) {
+    y.device(place) = x.prod(dim);
+  }
+};
+
+struct ProdGradFunctor {
+  template <typename DeviceContext, typename X, typename Y, typename DX,
+            typename DY, typename Dim>
+  void operator()(const DeviceContext& place, X& x, Y& y, DX& dx, DY& dy,
+                  const Dim& dim, int size) {
+    dx.device(place) = dy.broadcast(dim) * y.broadcast(dim) * x.inverse();
+  }
+};
+
 template <typename DeviceContext, typename T, typename Functor>
 class ReduceKernel : public framework::OpKernel<T> {
  public:
@@ -254,4 +270,5 @@ class ReduceGradKernel : public framework::OpKernel<T> {
   __macro(reduce_sum, SumFunctor, SumGradFunctor);      \
   __macro(reduce_mean, MeanFunctor, MeanGradFunctor);   \
   __macro(reduce_max, MaxFunctor, MaxOrMinGradFunctor); \
-  __macro(reduce_min, MinFunctor, MaxOrMinGradFunctor);
+  __macro(reduce_min, MinFunctor, MaxOrMinGradFunctor); \
+  __macro(reduce_prod, ProdFunctor, ProdGradFunctor);

python/paddle/fluid/layers/nn.py

@@ -49,6 +49,7 @@
     'reduce_mean',
     'reduce_max',
     'reduce_min',
+    'reduce_prod',
     'sequence_first_step',
     'sequence_last_step',
     'dropout',
@@ -2184,6 +2185,53 @@ def reduce_min(input, dim=None, keep_dim=False, name=None):
     return out
 
 
+def reduce_prod(input, dim=None, keep_dim=False, name=None):
+    """
+    Computes the product of tensor elements over the given dimension.
+
+    Args:
+        input (Variable): The input variable which is a Tensor or LoDTensor.
+        dim (int|None): The dimension along which the product is performed. If
+            :attr:`None`, multipy all elements of :attr:`input` and return a
+            Tensor variable with a single element, otherwise must be in the
+            range :math:`[-rank(input), rank(input))`. If :math:`dim < 0`,
+            the dimension to reduce is :math:`rank + dim`.
+        keep_dim (bool|False): Whether to reserve the reduced dimension in the
+            output Tensor. The result tensor will have one fewer dimension
+            than the :attr:`input` unless :attr:`keep_dim` is true.
+        name(str|None): A name for this layer(optional). If set None, the 
+            layer will be named automatically.
+
+    Returns:
+        Variable: The reduced Tensor variable.
+
+    Examples:
+        .. code-block:: python
+
+            # x is a Tensor variable with following elements:
+            #    [[0.2, 0.3, 0.5, 0.9]
+            #     [0.1, 0.2, 0.6, 0.7]]
+            # Each example is followed by the correspending output tensor.
+            fluid.layers.reduce_prod(x)  # [0.0002268]
+            fluid.layers.reduce_prod(x, dim=0)  # [0.02, 0.06, 0.3, 0.63]
+            fluid.layers.reduce_prod(x, dim=-1)  # [0.027, 0.0084]
+            fluid.layers.reduce_prod(x, dim=1, 
+                                     keep_dim=True)  # [[0.027], [0.0084]]
+    """
+    helper = LayerHelper('reduce_prod', **locals())
+    out = helper.create_tmp_variable(dtype=helper.input_dtype())
+    helper.append_op(
+        type='reduce_prod',
+        inputs={'X': input},
+        outputs={'Out': out},
+        attrs={
+            'dim': dim if dim != None else 0,
+            'keep_dim': keep_dim,
+            'reduce_all': True if dim == None else False
+        })
+    return out
+
+
 def split(input, num_or_sections, dim=-1, name=None):
     """
     Split the input tensor into multiple sub-tensors.

python/paddle/fluid/tests/unittests/test_reduce_op.py

@@ -70,6 +70,19 @@ def test_check_output(self):
         self.check_output()
 
 
+class TestProdOp(OpTest):
+    def setUp(self):
+        self.op_type = "reduce_prod"
+        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")}
+        self.outputs = {'Out': self.inputs['X'].prod(axis=0)}
+
+    def test_check_output(self):
+        self.check_output()
+
+    def test_check_grad(self):
+        self.check_grad(['X'], 'Out')
+
+
 class TestKeepDimReduce(OpTest):
     def setUp(self):
         self.op_type = "reduce_sum"