Polish nn code, test=develop (#22237)

* refine code, test=develop

* reuse contain_var, test=develop

python/paddle/fluid/framework.py

@@ -455,14 +455,14 @@ def name_scope(prefix=None):
     """
     # TODO(panyx0718): Only [0-9a-z].
     # in dygraph we don't need namescope since it will cause mem leak
-    if not in_dygraph_mode():
+    if in_dygraph_mode():
+        yield
+    else:
         assert prefix, "namescope prefix cannot be empty."
         global _name_scope
         _name_scope = _name_scope.child(prefix)
         yield
         _name_scope = _name_scope.parent()
-    else:
-        yield
 
 
 def _full_name_scope():
@@ -715,10 +715,9 @@ def _getitem_impl_(var, item):
     if (use_strided_slice == True):
         attrs['strides'] = []
     infer_flags = list(1 for i in range(len(slice_axis)))
+
     # starts
-    if not contain_var(slice_start):
-        attrs['starts'] = slice_start
-    else:
+    if contain_var(slice_start):
         inputs['StartsTensorList'] = get_new_list_tensor(slice_start)
         for i, dim in enumerate(slice_start):
             if isinstance(dim, Variable):
@@ -726,10 +725,11 @@ def _getitem_impl_(var, item):
                 infer_flags[i] = -1
             else:
                 attrs['starts'].append(dim)
-    # ends
-    if not contain_var(slice_end):
-        attrs['ends'] = slice_end
     else:
+        attrs['starts'] = slice_start
+
+    # ends
+    if contain_var(slice_end):
         inputs['EndsTensorList'] = get_new_list_tensor(slice_end)
         for i, dim in enumerate(slice_end):
             if isinstance(dim, Variable):
@@ -737,11 +737,12 @@ def _getitem_impl_(var, item):
                 infer_flags[i] = -1
             else:
                 attrs['ends'].append(dim)
+    else:
+        attrs['ends'] = slice_end
+
     # strides
     if use_strided_slice == True:
-        if not contain_var(slice_step):
-            attrs['strides'] = slice_step
-        else:
+        if contain_var(slice_step):
             inputs['StridesTensorList'] = get_new_list_tensor(slice_step)
             for i, dim in enumerate(slice_step):
                 if isinstance(dim, Variable):
@@ -749,6 +750,8 @@ def _getitem_impl_(var, item):
                     infer_flags[i] = -1
                 else:
                     attrs['strides'].append(dim)
+        else:
+            attrs['strides'] = slice_step
     # infer_flags
     attrs['infer_flags'] = infer_flags
 
@@ -2344,12 +2347,12 @@ class Block(object):
                 if isinstance(item[1], Parameter))
 
     def create_var(self, *args, **kwargs):
-        if not in_dygraph_mode():
+        if in_dygraph_mode():
+            var = _varbase_creator(*args, **kwargs)
+        else:
             var = Variable(block=self, *args, **kwargs)
             if 'initializer' in kwargs:
                 kwargs['initializer'](var, self)
-        else:
-            var = _varbase_creator(*args, **kwargs)
         return var
 
     def has_var(self, name):
@@ -2396,9 +2399,8 @@ class Block(object):
         # NOTE: v is destroyed by C++ after calling _rename_var.
         d = self.desc.find_var(cpt.to_bytes(new_name))
         if var_type == "Parameter":
-            if not in_dygraph_mode():
-                var = Parameter(
-                    self,
+            if in_dygraph_mode():
+                var = ParamBase(
                     d.shape(),
                     d.dtype(),
                     type=orig_var_type,
@@ -2410,7 +2412,8 @@ class Block(object):
                     gradient_clip_attr=gradient_clip_attr,
                     error_clip=error_clip)
             else:
-                var = ParamBase(
+                var = Parameter(
+                    self,
                     d.shape(),
                     d.dtype(),
                     type=orig_var_type,
@@ -2444,10 +2447,10 @@ class Block(object):
     def create_parameter(self, *args, **kwargs):
         global_block = self.program.global_block()
         param = None
-        if not in_dygraph_mode():
-            param = Parameter(global_block, *args, **kwargs)
-        else:
+        if in_dygraph_mode():
             param = ParamBase(*args, **kwargs)
+        else:
+            param = Parameter(global_block, *args, **kwargs)
         if 'initializer' in kwargs:
 
             def _is_inited_by(block, var):
@@ -2687,9 +2690,8 @@ class Block(object):
                                  "same topology")
             assert isinstance(v, Variable)
             new_p = None
-            if not in_dygraph_mode():
-                new_p = Parameter(
-                    block=self,
+            if in_dygraph_mode():
+                new_p = ParamBase(
                     shape=v.shape,
                     dtype=v.dtype,
                     type=v.type,
@@ -2702,7 +2704,8 @@ class Block(object):
                     error_clip=p.error_clip,
                     name=v.name)
             else:
-                new_p = ParamBase(
+                new_p = Parameter(
+                    block=self,
                     shape=v.shape,
                     dtype=v.dtype,
                     type=v.type,

python/paddle/fluid/layers/tensor.py

@@ -22,6 +22,7 @@ from ..initializer import Constant, force_init_on_cpu
 from ..core import VarDesc
 from .. import core
 from .layer_function_generator import templatedoc
+from . import utils
 from ..data_feeder import check_type_and_dtype, check_type, check_dtype, convert_dtype
 import numpy
 import warnings
@@ -552,13 +553,6 @@ def fill_constant(shape, dtype, value, force_cpu=False, out=None):
           shape = fluid.layers.fill_constant([1,2], "int32", 2) # shape=[2,2]
           data4 = fluid.layers.fill_constant(shape=shape, dtype='bool', value=True) # data4=[[True,True],[True,True]]
     """
-
-    def _contain_var(one_list):
-        for ele in one_list:
-            if isinstance(ele, Variable):
-                return True
-        return False
-
     attrs = {
         'value': float(value),
         'force_cpu': force_cpu or force_init_on_cpu()
@@ -571,8 +565,7 @@ def fill_constant(shape, dtype, value, force_cpu=False, out=None):
 
     if in_dygraph_mode():
         if isinstance(shape, (list, tuple)):
-            contain_var = _contain_var(shape)
-            if contain_var:
+            if utils._contain_var(shape):
                 raise TypeError(
                     "The type of 'shape' in fill_constant must be list[int] or tuple(int) in Dygraph mode, but "
                     "received %s, which contains Variable." % type(shape))
@@ -644,7 +637,7 @@ def fill_constant(shape, dtype, value, force_cpu=False, out=None):
             "The size of 'shape' in fill_constant can't be zero, "
             "but received %s." % len(shape))
         attrs["shape"] = _get_attr_shape(shape)
-        if _contain_var(shape):
+        if utils._contain_var(shape):
             inputs['ShapeTensorList'] = _get_shape_tensor(shape)
 
     if out is None:

python/paddle/fluid/layers/utils.py

@@ -16,6 +16,7 @@ from __future__ import print_function
 import collections
 import six
 import numpy as np
+from ..framework import Variable
 
 
 def convert_to_list(value, n, name, dtype=np.int):
@@ -244,3 +245,13 @@ def _is_symmetric_padding(padding, data_dim):
             if padding[i * 2] != padding[i * 2 + 1]:
                 is_sys = False
     return is_sys
+
+
+def _contain_var(list_or_tuple):
+    """
+    Check whether list or tuple contains variable.
+    """
+    for item in list_or_tuple:
+        if isinstance(item, Variable):
+            return True
+    return False

python/paddle/fluid/optimizer.py

@@ -134,11 +134,11 @@ class Optimizer(object):
         # global step if use lr decay
         if isinstance(self._learning_rate, LearningRateDecay):
             var_tmp = None
-            if not framework.in_dygraph_mode():
-                var_temp = Variable(None, name='global_step', dtype='int32')
-            else:
+            if framework.in_dygraph_mode():
                 var_temp = framework._varbase_creator(
                     None, name='global_step', dtype='int32')
+            else:
+                var_temp = Variable(None, name='global_step', dtype='int32')
 
             tensor.fill_constant(
                 [1], "int32", self._learning_rate.step_num, out=var_temp)
@@ -546,10 +546,10 @@ class Optimizer(object):
             See examples in ``apply_gradients``.
         """
         act_no_grad_set = None
-        if not framework.in_dygraph_mode():
-            act_no_grad_set = self._get_no_grad_set(loss, no_grad_set)
-        else:
+        if framework.in_dygraph_mode():
             pass
+        else:
+            act_no_grad_set = self._get_no_grad_set(loss, no_grad_set)
 
         self._dtype = loss.dtype
         if framework.in_dygraph_mode():