rand API: remove out, device, stop_gradient; add name

paddle/fluid/operators/gaussian_random_op.cc

@@ -98,7 +98,7 @@ class GaussianRandomOp : public framework::OperatorWithKernel {
 
       return;
     }
-    if (!(ctx->HasInput("ShapeTensor") && !ctx->HasInputs("ShapeTensorList"))) {
+    if (!ctx->HasInput("ShapeTensor") && !ctx->HasInputs("ShapeTensorList")) {
       PADDLE_ENFORCE_GT(
           shape.size(), 0UL,
           platform::errors::InvalidArgument(

python/paddle/fluid/layers/nn.py

@@ -10486,29 +10486,24 @@ def gaussian_random(shape, mean=0.0, std=1.0, seed=0, dtype='float32'):
            #        [2.8675377 , 2.2279181 , 0.79029655, 2.8447366 ]], dtype=float32)
     """
 
-    helper = LayerHelper('gaussian_random', **locals())
-    out = helper.create_variable_for_type_inference(dtype)
-    if not isinstance(shape, (list, tuple, Variable)):
-        raise TypeError(
-            "The type of 'shape' in fill_constant must be Variable, list or tuple, but "
-            "received %s." % (type(shape)))
-    c_dtype = convert_np_dtype_to_dtype_(dtype)
+    check_type(shape, 'shape', (list, tuple, Variable), 'gaussian_random')
+    if not isinstance(dtype, core.VarDesc.VarType):
+        dtype = convert_np_dtype_to_dtype_(dtype)
+    check_dtype(dtype, 'dtype', ['float32', 'float64'], 'gaussian_random')
+
+    inputs = {}
     attrs = {
         'mean': mean,
         'std': std,
         'seed': seed,
-        'dtype': c_dtype,
+        'dtype': dtype,
         'use_mkldnn': False
     }
-
-    inputs = {}
     utils._get_shape_tensor_inputs(
-        inputs=inputs,
-        helper=helper,
-        attrs=attrs,
-        shape=shape,
-        op_type='gaussian_random')
+        inputs=inputs, attrs=attrs, shape=shape, op_type='gaussian_random')
 
+    helper = LayerHelper('gaussian_random', **locals())
+    out = helper.create_variable_for_type_inference(dtype)
     helper.append_op(
         type='gaussian_random',
         inputs=inputs,
@@ -14861,7 +14856,8 @@ def gather_tree(ids, parents):
 
 
 @templatedoc()
-def uniform_random(shape, dtype='float32', min=-1.0, max=1.0, seed=0):
+def uniform_random(shape, dtype='float32', min=-1.0, max=1.0, seed=0,
+                   name=None):
     """
     This OP initializes a variable with random values sampled from a
     uniform distribution in the range [min, max).
@@ -14876,18 +14872,24 @@ def uniform_random(shape, dtype='float32', min=-1.0, max=1.0, seed=0):
           result=[[0.8505902, 0.8397286]]
 
     Args:
-        shape (list|tuple|Variable): The shape of the output Tensor,  if the shape is a list or tuple,
-                                     its elements can be an integer
-                                     or a Tensor with the shape [1], and the type of the Tensor must be int32 or int64.
-                                     If the shape is a Variable, it is a 1-D Tensor, and the type of the Tensor must be int32 or int64.
-        dtype(np.dtype|core.VarDesc.VarType|str, optional): The type of the output Tensor. Supported data types: float32, float64.
-                                                  Default: float32.
-        min (float, optional): The lower bound on the range of random values to generate, the min is included in the range. Default -1.0.
-        max (float, optional): The upper bound on the range of random values to generate, the max is excluded in the range. Default 1.0.
-        seed (int, optional): Random seed used for generating samples. 0 means use a
-            seed generated by the system. Note that if seed is not 0, this
-            operator will always generate the same random numbers every time.
-            Default 0.
+        shape (list|tuple|Variable): The shape of the output Tensor,  if the
+            shape is a list or tuple, its elements can be an integer or a
+            Tensor with the shape [1], and the type of the Tensor must be
+            int32 or int64. If the shape is a Variable, it is a 1-D Tensor, and
+            the type of the Tensor must be int32 or int64.
+        dtype(np.dtype|core.VarDesc.VarType|str, optional): The type of the
+            output Tensor. Supported data types: float32, float64. Default: float32.
+        min (float, optional): The lower bound on the range of random values
+            to generate, the min is included in the range. Default -1.0.
+        max (float, optional): The upper bound on the range of random values
+            to generate, the max is excluded in the range. Default 1.0.
+        seed (int, optional): Random seed used for generating samples. 0 means
+            use a seed generated by the system. Note that if seed is not 0,
+            this operator will always generate the same random numbers every
+            time. Default 0.
+        name(str, optional): The default value is None. Normally there is no
+            need for user to set this property. For more information, please
+            refer to :ref:`api_guide_Name`.
 
     Returns:
         Variable: A Tensor of the specified shape filled with uniform_random values.
@@ -14917,62 +14919,30 @@ def uniform_random(shape, dtype='float32', min=-1.0, max=1.0, seed=0):
             var_shape_int32 = fluid.data(name='var_shape_int32', shape=[2], dtype="int32")
             result_4 = fluid.layers.uniform_random(var_shape_int32)
 
-
-
     """
-    check_type(shape, 'shape', (list, tuple, Variable), 'uniform_random')
     if not isinstance(dtype, core.VarDesc.VarType):
         dtype = convert_np_dtype_to_dtype_(dtype)
-    check_dtype(dtype, 'dtype', ('float32', 'float64'), 'uniform_random')
 
-    def get_new_shape_tensor(list_shape):
-        new_shape_tensor = []
-        for dim in list_shape:
-            if isinstance(dim, Variable):
-                dim.stop_gradient = True
-                new_shape_tensor.append(dim)
-            else:
-                assert (isinstance(dim, int))
-                temp_out = helper.create_variable_for_type_inference('int64')
-                fill_constant([1], 'int64', dim, force_cpu=True, out=temp_out)
-                new_shape_tensor.append(temp_out)
-        return new_shape_tensor
+    if in_dygraph_mode():
+        shape = utils._convert_shape_to_list(shape)
+        return core.ops.uniform_random('shape', shape, 'min',
+                                       float(min), 'max',
+                                       float(max), 'seed', seed, 'dtype', dtype)
 
-    def get_attr_shape(list_shape):
-        unk_dim_idx = -1
-        attrs_shape = []
-        for dim_idx, dim_size in enumerate(list_shape):
-            if isinstance(dim_size, Variable):
-                attrs_shape.append(-1)
-            else:
-                attrs_shape.append(dim_size)
-                assert dim_size > 0, (
-                    "Each dimension size given in shape must not be negative "
-                    "except one unknown dimension.")
-        return attrs_shape
+    check_type(shape, 'shape', (list, tuple, Variable), 'uniform_random')
+    check_dtype(dtype, 'dtype', ('float32', 'float64'), 'uniform_random')
 
-    helper = LayerHelper("uniform_random", **locals())
     inputs = dict()
     attrs = {'seed': seed, 'min': min, 'max': max, 'dtype': dtype}
-    if in_dygraph_mode():
-        attrs['shape'] = shape
-    else:
-        if isinstance(shape, Variable):
-            shape.stop_gradient = True
-            inputs["ShapeTensor"] = shape
-        elif isinstance(shape, (list, tuple)):
-            assert len(shape) > 0, (
-                "The size of argument(shape) can't be zero.")
-            attrs["shape"] = get_attr_shape(shape)
-            if utils._contain_var(shape):
-                inputs['ShapeTensorList'] = get_new_shape_tensor(shape)
+    utils._get_shape_tensor_inputs(
+        inputs=inputs, attrs=attrs, shape=shape, op_type='uniform_random')
 
+    helper = LayerHelper("uniform_random", **locals())
     out = helper.create_variable_for_type_inference(dtype)
     helper.append_op(
         type="uniform_random", inputs=inputs, attrs=attrs,
         outputs={"Out": out})
-
-    return helper.append_activation(out)
+    return out
 
 
 def unbind(input, axis=0):

python/paddle/fluid/layers/tensor.py

@@ -685,12 +685,7 @@ def fill_constant(shape, dtype, value, force_cpu=False, out=None):
             attrs['str_value'] = str(float(value))
 
     if in_dygraph_mode():
-        if isinstance(shape, (list, tuple)):
-            shape = list(
-                map(lambda x: x.numpy()[0] if isinstance(x, Variable) else x,
-                    shape))
-        else:
-            shape = list(shape.numpy().astype(int))
+        shape = utils._convert_shape_to_list(shape)
         if out is None:
             out = _varbase_creator(dtype=dtype)
 
@@ -719,12 +714,8 @@ def fill_constant(shape, dtype, value, force_cpu=False, out=None):
                                  'fill_constant')
 
     helper = LayerHelper("fill_constant", **locals())
-    inputs = utils._get_shape_tensor_inputs(
-        inputs=inputs,
-        helper=helper,
-        attrs=attrs,
-        shape=shape,
-        op_type='fill_constant')
+    utils._get_shape_tensor_inputs(
+        inputs=inputs, attrs=attrs, shape=shape, op_type='fill_constant')
 
     if out is None:
         out = helper.create_variable_for_type_inference(dtype=dtype)

python/paddle/fluid/layers/utils.py

@@ -282,7 +282,7 @@ def _contain_var(list_or_tuple):
     return False
 
 
-def _get_shape_tensor_inputs(inputs, helper, attrs, shape, op_type):
+def _get_shape_tensor_inputs(inputs, attrs, shape, op_type):
     from .tensor import fill_constant, cast
 
     def _get_attr_shape(list_shape):
@@ -295,7 +295,7 @@ def _get_attr_shape(list_shape):
         return attr_shape
 
     def _get_shape_tensor(list_shape):
-        new_shape_tensor = []
+        shape_tensor_list = []
         for idx, dim in enumerate(list_shape):
             if isinstance(dim, Variable):
                 dim.stop_gradient = True
@@ -305,11 +305,11 @@ def _get_shape_tensor(list_shape):
                     '(When type of shape in' + op_type + 'is list or tuple.)')
                 if convert_dtype(dim.dtype) == 'int64':
                     dim = cast(x=dim, dtype='int32')
-                new_shape_tensor.append(dim)
+                shape_tensor_list.append(dim)
             else:
                 temp_out = fill_constant([1], 'int32', dim, force_cpu=True)
-                new_shape_tensor.append(temp_out)
-        return new_shape_tensor
+                shape_tensor_list.append(temp_out)
+        return shape_tensor_list
 
     if isinstance(shape, Variable):
         shape.stop_gradient = True
@@ -325,8 +325,8 @@ def _get_shape_tensor(list_shape):
         attrs["shape"] = _get_attr_shape(shape)
         if _contain_var(shape):
             inputs['ShapeTensorList'] = _get_shape_tensor(shape)
-
-    return inputs
+    else:
+        raise TypeError("Shape only supports Variable, or list, or tuple.")
 
 
 def _convert_to_tensor_list(old_list, dtype="int32"):
@@ -345,3 +345,16 @@ def _convert_to_tensor_list(old_list, dtype="int32"):
             temp_out = fill_constant([1], dtype, ele, force_cpu=True)
             new_list_tensor.append(temp_out)
     return new_list_tensor
+
+
+def _convert_shape_to_list(shape):
+    """
+    Convert shape(list, tuple, variable) to list in imperative mode
+    """
+    if isinstance(shape, (list, tuple)):
+        shape = list(
+            map(lambda x: x.numpy()[0] if isinstance(x, Variable) else x,
+                shape))
+    else:
+        shape = list(shape.numpy().astype(int))
+    return shape

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

@@ -47,71 +47,73 @@ def test_dtype():
 
             self.assertRaises(TypeError, test_dtype)
 
-            def test_shape_list():
-                rand(shape=[2.])
-
-            self.assertRaises(TypeError, test_shape_list)
-
-            def test_shape_list2():
-                rand(shape=[2, 3.])
-
-            self.assertRaises(TypeError, test_shape_list2)
-
-            def test_device():
-                rand(shape=[3, 4], device='device')
-
-            self.assertRaises(ValueError, test_device)
-
 
 class TestRandOp(unittest.TestCase):
     """
     This class test the common usages of randop.
-
     """
 
-    def test_run(self):
-        use_cuda = False
+    def run_net(self, use_cuda=False):
         place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()
         exe = fluid.Executor(place)
 
         train_program = fluid.Program()
         startup_program = fluid.Program()
         with fluid.program_guard(train_program, startup_program):
-            result_1 = rand(shape=[3, 4])
+            result_0 = rand([3, 4])
+            result_1 = rand([3, 4], 'float64')
+
             dim_1 = fluid.layers.fill_constant([1], "int64", 3)
             dim_2 = fluid.layers.fill_constant([1], "int32", 5)
             result_2 = rand(shape=[dim_1, dim_2])
+
             var_shape = fluid.data(name='var_shape', shape=[2], dtype="int64")
             result_3 = rand(var_shape)
+
             var_shape_int32 = fluid.data(
                 name='var_shape_int32', shape=[2], dtype="int32")
             result_4 = rand(var_shape_int32)
+
         exe.run(startup_program)
 
         x1 = np.array([3, 2]).astype('int64')
         x2 = np.array([4, 3]).astype('int32')
-        ret = exe.run(train_program,
-                      feed={"var_shape": x1,
-                            "var_shape_int32": x2},
-                      fetch_list=[result_1, result_2, result_3, result_4])
+        ret = exe.run(
+            train_program,
+            feed={"var_shape": x1,
+                  "var_shape_int32": x2},
+            fetch_list=[result_1, result_1, result_2, result_3, result_4])
+
+    def test_run(self):
+        self.run_net(False)
+        if core.is_compiled_with_cuda():
+            self.run_net(True)
 
 
 class TestRandOpForDygraph(unittest.TestCase):
     """
     This class test the common usages of randop.
-
     """
 
-    def test_run(self):
-        use_cuda = False
-        with fluid.dygraph.guard():
-            rand(shape=[3, 4])
+    def run_net(self, use_cuda=False):
+        place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()
+        with fluid.dygraph.guard(place):
+            rand([3, 4])
+
+            rand([3, 4], 'float64')
+
             dim_1 = fluid.layers.fill_constant([1], "int64", 3)
             dim_2 = fluid.layers.fill_constant([1], "int32", 5)
             rand(shape=[dim_1, dim_2])
+
             var_shape = fluid.dygraph.to_variable(np.array([3, 4]))
             rand(var_shape)
 
+    def test_run(self):
+        self.run_net(False)
+        if core.is_compiled_with_cuda():
+            self.run_net(True)
+
 
 if __name__ == "__main__":
     unittest.main()