Merge branch 'develop' into api

benchmark/fluid/fluid_benchmark.py

@@ -210,7 +210,7 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
     # generate fake:
     if args.use_fake_data:
         for var in feed_var_list:
-            v = startup_prog.global_block().clone_variable(var)
+            v = startup_prog.global_block()._clone_variable(var)
             var.persistable = True
             v.persistable = True
 

doc/fluid/design/modules/python_api.md

@@ -98,13 +98,13 @@ class Block(objects):
     def append_operator(self, ...):
         self.ops.append(Operator(self, ...))
 
-    def prepend_operator(self, ...): # Parameter's ctor prepands initialize operators.
+    def _prepend_operator(self, ...): # Parameter's ctor prepands initialize operators.
        self.ops.prepend(Operator(self, ...))
 ```
 
 `create_parameter` is necessary because parameters are global variables, defined in the global block, but can be created in some sub-blocks. For example, an FC layer in the step block of an RNN operator.
 
-`prepend_operator` is necessary because the constructor of `Parameter` needs to create the initialize (or load) operator of the parameter, and would like to put it in the *preamble* of the global block.
+`_prepend_operator` is necessary because the constructor of `Parameter` needs to create the initialize (or load) operator of the parameter, and would like to put it in the *preamble* of the global block.
 
 ### Operator
 

doc/fluid/howto/performance/error_clip.md

@@ -78,7 +78,7 @@ def error_clip_callback(block, context):
     op_desc = block.desc.op(block.desc.op_size() - 1)
     for grad_n in filter(lambda n: grad_to_var.has_key(n),
                          op_desc.output_arg_names()):
-        fwd_var = block.var_recursive(grad_to_var[grad_n])
+        fwd_var = block.__var_recursive(grad_to_var[grad_n])
         error_clip = getattr(fwd_var, "error_clip", None)
         if not (error_clip is None or isinstance(error_clip,
                                                  BaseErrorClipAttr)):

doc/v2/build_and_install/build_from_source_cn.rst

@@ -35,11 +35,16 @@ PaddlePaddle需要使用Docker环境完成编译，这样可以免去单独安
    # 2. 可选步骤：源码中构建用于编译PaddlePaddle的Docker镜像
    docker build -t paddle:dev .
    # 3. 执行下面的命令编译CPU-Only的二进制
-   docker run -it -v $PWD:/paddle -w /paddle -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddlepaddle/paddle_manylinux_devel:cuda8.0_cudnn5 ./paddle/scripts/paddle_build.sh build
+   docker run -it -v $PWD:/paddle -w /paddle -e "PYTHON_ABI=cp27-cp27mu" -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddlepaddle/paddle_manylinux_devel:cuda8.0_cudnn5 ./paddle/scripts/paddle_build.sh build
    # 4. 或者也可以使用为上述可选步骤构建的镜像（必须先执行第2步）
    docker run -it -v $PWD:/paddle -w /paddle -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddle:dev ./paddle/scripts/paddle_build.sh build
 
-注：上述命令把当前目录（源码树根目录）映射为 container 里的 :code:`/paddle` 目录。
+注：
+
+- 上述命令把当前目录（源码树根目录）映射为 container 里的 :code:`/paddle` 目录。
+
+- 如果您使用的是 manylinux 的镜像进行编译, 那么您需要通过环境变量 :code:`PYTHON_ABI` 来指定一个 `Python ABI <https://www.python.org/dev/peps/pep-0425/#id8>`__.
+PaddlePaddle目前支持的 Python ABI 有 :code:`cp27-cp27m` 和 :code:`cp27-cp27mu`.
 
 编译完成后会在build/python/dist目录下生成输出的whl包，可以选在在当前机器安装也可以拷贝到目标机器安装：
 

doc/v2/build_and_install/build_from_source_en.rst

@@ -36,13 +36,18 @@ If you don't wish to use docker，you need to install several compile dependenci
    # 2. Optional: build development docker image from source
    docker build -t paddle:dev .
    # 3. Run the following command to build a CPU-Only binaries
-   docker run -it -v $PWD:/paddle -w /paddle -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddlepaddle/paddle_manylinux_devel:cuda8.0_cudnn5 ./paddle/scripts/paddle_build.sh build
+   docker run -it -v $PWD:/paddle -w /paddle -e "PYTHON_ABI=cp27-cp27mu" -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddlepaddle/paddle_manylinux_devel:cuda8.0_cudnn5 ./paddle/scripts/paddle_build.sh build
    # 4. Or, use your built Docker image to build PaddlePaddle (must run step 2)
    docker run -it -v $PWD:/paddle -w /paddle -e "WITH_GPU=OFF" -e "WITH_TESTING=OFF" paddle:dev ./paddle/scripts/paddle_build.sh build
 
-NOTE: The above command try to mount the current working directory (root directory of source code)
+NOTE: 
+
+- The above command try to mount the current working directory (root directory of source code)
 into :code:`/paddle` directory inside docker container.
 
+- You need to pass in the required environment variable :code:`PYTHON_ABI` to specify a `Python ABI <https://www.python.org/dev/peps/pep-0425/#id8>`__.
+Currently PaddlePaddle supported Python ABIs include :code:`cp27-cp27m` and :code:`cp27-cp27mu` .
+
 When the compile finishes, you can get the output whl package under
 build/python/dist, then you can choose to install the whl on local
 machine or copy it to the target machine.

paddle/contrib/float16/float16_transpiler.py

@@ -118,7 +118,7 @@ class Float16Transpiler:
 
         for var in self.block.vars.keys():
             if var not in args:
-                self.block.remove_var(var)
+                self.block._remove_var(var)
 
     def _modify_feed_fetch(self):
         '''
@@ -165,7 +165,7 @@ class Float16Transpiler:
                     dtype=core.VarDesc.VarType.FP16,
                     shape=var.shape,
                     persistable=var.persistable)
-                self.block.insert_op(
+                self.block._insert_op(
                     i + 1,
                     type="cast",
                     inputs={"X": var},
@@ -188,7 +188,7 @@ class Float16Transpiler:
                     persistable=var.persistable)
                 find_op(var)
                 var.op.rename_output(var_name, tmp_var_name)
-                self.block.insert_op(
+                self.block._insert_op(
                     i,
                     type="cast",
                     inputs={"X": tmp_var},
@@ -253,4 +253,4 @@ class Float16Transpiler:
 
             # old var will be replaced by the fp16 var in program desc
             self.input_map[var.name] = fp16_var_name
-            self.block.remove_var(var.name)
+            self.block._remove_var(var.name)

paddle/fluid/pybind/protobuf.cc

@@ -145,14 +145,14 @@ void BindBlockDesc(pybind11::module *m) {
       .def_property_readonly("id", &pd::BlockDesc::ID)
       .def_property_readonly("parent", &pd::BlockDesc::Parent)
       .def("get_forward_block_idx", &pd::BlockDesc::ForwardBlockID)
-      .def("set_forward_block_idx", &pd::BlockDesc::SetForwardBlockID)
+      .def("_set_forward_block_idx", &pd::BlockDesc::SetForwardBlockID)
       .def("append_op", &pd::BlockDesc::AppendOp,
            pybind11::return_value_policy::reference)
-      .def("prepend_op", &pd::BlockDesc::PrependOp,
+      .def("_prepend_op", &pd::BlockDesc::PrependOp,
            pybind11::return_value_policy::reference)
-      .def("insert_op", &pd::BlockDesc::InsertOp,
+      .def("_insert_op", &pd::BlockDesc::InsertOp,
            pybind11::return_value_policy::reference)
-      .def("remove_op", &pd::BlockDesc::RemoveOp)
+      .def("_remove_op", &pd::BlockDesc::RemoveOp)
       .def("var",
            [](pd::BlockDesc &self, pybind11::bytes byte_name) {
              std::string name = byte_name;
@@ -165,7 +165,7 @@ void BindBlockDesc(pybind11::module *m) {
              return self.HasVar(name);
            },
            pybind11::return_value_policy::reference)
-      .def("rename_var",
+      .def("_rename_var",
            [](pd::BlockDesc &self, const pybind11::bytes &byte_name,
               const pybind11::bytes &byte_name_new) {
              std::string name = byte_name;
@@ -189,7 +189,7 @@ void BindBlockDesc(pybind11::module *m) {
              return self.FindVarRecursive(name);
            },
            pybind11::return_value_policy::reference)
-      .def("remove_var",
+      .def("_remove_var",
            [](pd::BlockDesc &self, pybind11::bytes byte_name) {
              std::string name = byte_name;
              return self.RemoveVar(name);

python/paddle/fluid/__init__.py

@@ -56,6 +56,7 @@ import unique_name
 import recordio_writer
 import parallel_executor
 from parallel_executor import *
+from paddle.fluid.layers.math_op_patch import monkey_patch_variable
 
 Tensor = LoDTensor
 
@@ -138,5 +139,5 @@ def __bootstrap__():
 
 # TODO(panyx0718): Avoid doing complex initialization logic in __init__.py.
 # Consider paddle.init(args) or paddle.main(args)
-layers.monkey_patch_variable()
+monkey_patch_variable()
 __bootstrap__()

python/paddle/fluid/backward.py

@@ -328,7 +328,7 @@ def _append_backward_ops_(block,
         if op.has_attr("sub_block"):
             sub_block = program.block(op.block_attr("sub_block"))
             grad_sub_block = program.create_block()
-            grad_sub_block.set_forward_block_idx(sub_block.idx)
+            grad_sub_block._set_forward_block_idx(sub_block.idx)
             cb = _callback_lookup_(op)
             if cb is not None:
                 if callbacks is None:
@@ -571,7 +571,7 @@ def append_backward(loss, parameter_list=None, no_grad_set=None,
     _append_backward_vars_(root_block, fwd_op_num, grad_to_var, grad_info_map)
 
     program.current_block_idx = current_block_idx
-    program.sync_with_cpp()
+    program._sync_with_cpp()
     # FIXME(zcd): prevent loss.grad optimized by mem_opt.
     loss.block.var(_append_grad_suffix_(loss.name)).persistable = True
 
@@ -744,7 +744,7 @@ def calc_gradient(targets, inputs, target_gradients=None, no_grad_set=None):
     _rename_grad_(block, fwd_op_num, grad_to_var, target_grad_map)
 
     _append_backward_vars_(block, fwd_op_num, grad_to_var, grad_info_map)
-    prog.sync_with_cpp()
+    prog._sync_with_cpp()
 
     grad_vars = []
     for input_var in inputs:

python/paddle/fluid/clip.py

@@ -82,7 +82,7 @@ def error_clip_callback(block, context):
     op_desc = block.desc.op(block.desc.op_size() - 1)
     for grad_n in filter(lambda n: grad_to_var.has_key(n),
                          op_desc.output_arg_names()):
-        fwd_var = block.var_recursive(grad_to_var[grad_n])
+        fwd_var = block._var_recursive(grad_to_var[grad_n])
         error_clip = getattr(fwd_var, "error_clip", None)
         if not (error_clip is None or isinstance(error_clip,
                                                  BaseErrorClipAttr)):

python/paddle/fluid/concurrency.py

@@ -69,8 +69,10 @@ class Go(BlockGuard):
         parent_block.append_op(
             type='go',
             inputs={
-                'X':
-                [parent_block.var_recursive(x_name) for x_name in x_name_list]
+                'X': [
+                    parent_block._var_recursive(x_name)
+                    for x_name in x_name_list
+                ]
             },
             outputs={},
             attrs={'sub_block': go_block})
@@ -259,7 +261,7 @@ class Select(BlockGuard):
             if var_name in intermediate
         ]
 
-        X = [select_block.var_recursive(x_name) for x_name in params]
+        X = [select_block._var_recursive(x_name) for x_name in params]
 
         # Needs to be used by `equal` inside the cases block.
         X.append(self.case_to_execute)

python/paddle/fluid/executor.py

@@ -309,7 +309,7 @@ class Executor(object):
         if not has_feed_operators(global_block, feed, feed_var_name):
             for i, name in enumerate(feed):
                 out = global_block.var(name)
-                global_block.prepend_op(
+                global_block._prepend_op(
                     type='feed',
                     inputs={'X': [feed_var]},
                     outputs={'Out': [out]},

python/paddle/fluid/framework.py

@@ -32,7 +32,6 @@ except Exception, e:
 import unique_name
 
 __all__ = [
-    'Block',
     'Variable',
     'Program',
     'Operator',
@@ -447,7 +446,7 @@ class Operator(object):
 
     Notes:
         The constructor of operator should not be invoked directly. Use
-        Block.append_op or Block.prepend_op instead.
+        Block.append_op or Block._prepend_op instead.
 
     Examples:
         .. code-block:: python
@@ -870,7 +869,7 @@ class Block(object):
     def forward_block_idx(self):
         return self.desc.get_forward_block_idx()
 
-    def set_forward_block_idx(self, idx):
+    def _set_forward_block_idx(self, idx):
         """
         Set the forward block Idx.
 
@@ -880,7 +879,7 @@ class Block(object):
         Returns:
             None
         """
-        self.desc.set_forward_block_idx(idx)
+        self.desc._set_forward_block_idx(idx)
 
     @property
     def idx(self):
@@ -909,7 +908,7 @@ class Block(object):
             raise ValueError("var %s not in this block" % name)
         return v
 
-    def var_recursive(self, name):
+    def _var_recursive(self, name):
         """
         Get a Variable by name from this block recursively.
 
@@ -951,9 +950,9 @@ class Block(object):
         raise ValueError("Var {0} is not found recursively".format(name))
 
     def all_parameters(self):
-        return list(self.iter_parameters())
+        return list(self._iter_parameters())
 
-    def iter_parameters(self):
+    def _iter_parameters(self):
         return (item[1] for item in self.vars.iteritems()
                 if isinstance(item[1], Parameter))
 
@@ -966,7 +965,7 @@ class Block(object):
     def has_var(self, name):
         return name in self.vars
 
-    def rename_var(self, name, new_name):
+    def _rename_var(self, name, new_name):
         """
         Rename variable in vars and ops' inputs and outputs
 
@@ -1000,8 +999,8 @@ class Block(object):
         else:
             raise ValueError("unsupported var type: %s", type(v))
         orig_var_type = v.type
-        self.desc.rename_var(name, new_name)
-        # NOTE: v is destroyed by C++ after calling rename_var.
+        self.desc._rename_var(name, new_name)
+        # NOTE: v is destroyed by C++ after calling _rename_var.
         d = self.desc.find_var(new_name)
         if var_type == "Parameter":
             var = Parameter(
@@ -1024,16 +1023,16 @@ class Block(object):
                 error_clip=error_clip,
                 stop_gradient=stop_gradient)
 
-        # rename the python side, sync_with_cpp will only add
+        # rename the python side, _sync_with_cpp will only add
         # new vars/ops to python side.
         self.vars[new_name] = var
         del self.vars[name]
-        self.sync_with_cpp()
+        self._sync_with_cpp()
         return var
 
-    def remove_var(self, name):
-        self.sync_with_cpp()
-        self.desc.remove_var(name)
+    def _remove_var(self, name):
+        self._sync_with_cpp()
+        self.desc._remove_var(name)
         del self.vars[name]
 
     def create_parameter(self, *args, **kwargs):
@@ -1055,7 +1054,7 @@ class Block(object):
         self.ops.append(op)
         return op
 
-    def insert_op(self, index, *args, **kwargs):
+    def _insert_op(self, index, *args, **kwargs):
         """
         Insert a Operator according to the giving arguments.
 
@@ -1065,13 +1064,13 @@ class Block(object):
         Returns:
             Operator: the insert Operator.
         """
-        self.sync_with_cpp()
-        op_desc = self.desc.insert_op(index)
+        self._sync_with_cpp()
+        op_desc = self.desc._insert_op(index)
         op = Operator(block=self, desc=op_desc, *args, **kwargs)
         self.ops.insert(index, op)
         return op
 
-    def remove_op(self, index):
+    def _remove_op(self, index):
         """
         Remove the specific position operator.
 
@@ -1081,11 +1080,11 @@ class Block(object):
         Returns:
             None
         """
-        self.sync_with_cpp()
-        self.desc.remove_op(index, index + 1)
+        self._sync_with_cpp()
+        self.desc._remove_op(index, index + 1)
         del self.ops[index]
 
-    def slice_ops(self, start, end):
+    def _slice_ops(self, start, end):
         """
         Return the Operator between start and end.
 
@@ -1098,13 +1097,13 @@ class Block(object):
         """
         return self.ops[start:end]
 
-    def prepend_op(self, *args, **kwargs):
-        op_desc = self.desc.prepend_op()
+    def _prepend_op(self, *args, **kwargs):
+        op_desc = self.desc._prepend_op()
         op = Operator(self, op_desc, *args, **kwargs)
         self.ops.insert(0, op)
         return op
 
-    def sync_with_cpp(self):
+    def _sync_with_cpp(self):
         """
         Sync from the desc on the c++ end. This method is used to synchronize
         the c++ desc instance generated by backward.
@@ -1170,7 +1169,7 @@ class Block(object):
         for index in range(len(self.ops)):
             assert self.ops[index].desc == ops_in_cpp[index]
 
-    def copy_param_info_from(self, other):
+    def _copy_param_info_from(self, other):
         """
         Copy the information of parameters from the other block.
 
@@ -1185,12 +1184,13 @@ class Block(object):
             None
         """
         if not isinstance(other, Block):
-            raise TypeError("copy_param_info_from should be invoked with Block")
-        for p in other.iter_parameters():
+            raise TypeError(
+                "_copy_param_info_from should be invoked with Block")
+        for p in other._iter_parameters():
             assert isinstance(p, Parameter)
             v = self.vars.get(p.name, None)
             if v is None:
-                raise ValueError("copy_param_info_from should be invoked with "
+                raise ValueError("_copy_param_info_from should be invoked with "
                                  "same topology")
             assert isinstance(v, Variable)
             new_p = Parameter(
@@ -1208,7 +1208,7 @@ class Block(object):
                 name=v.name)
             self.vars[new_p.name] = new_p
 
-    def clone_variable(self, var):
+    def _clone_variable(self, var):
         """
         Clone a variable into current block.
 
@@ -1484,9 +1484,9 @@ class Program(object):
             p = Program()
             p.desc = core.ProgramDesc(self.desc)
             p.blocks = [Block(p, i) for i in xrange(self.desc.num_blocks())]
-            p.sync_with_cpp()
+            p._sync_with_cpp()
 
-        p.copy_param_info_from(self)
+        p._copy_param_info_from(self)
         p.copy_data_info_from(self)
         return p
 
@@ -1536,7 +1536,7 @@ class Program(object):
         res = Program()
         res.desc = core.prune(self.desc, targets_idx)
         res.blocks = [Block(res, i) for i in xrange(res.desc.num_blocks())]
-        res.sync_with_cpp()
+        res._sync_with_cpp()
         return res
 
     def inference_optimize(self):
@@ -1562,7 +1562,7 @@ class Program(object):
                 if op.has_attr('is_test'):
                     op.set_attr('is_test', True)
         res.blocks = [Block(res, i) for i in xrange(res.desc.num_blocks())]
-        res.sync_with_cpp()
+        res._sync_with_cpp()
         return res
 
     @staticmethod
@@ -1582,7 +1582,7 @@ class Program(object):
         p = Program()
         p.desc = core.ProgramDesc(binary_str)
         p.blocks = [Block(p, i) for i in xrange(p.desc.num_blocks())]
-        p.sync_with_cpp()
+        p._sync_with_cpp()
         return p
 
     @property
@@ -1662,7 +1662,7 @@ class Program(object):
         """
         self.current_block_idx = self.current_block().parent_idx
 
-    def sync_with_cpp(self):
+    def _sync_with_cpp(self):
         """
         Synchronize Python instance to its binding C++ object instance.
         If the program is modified in C++ space, this method should be invoked.
@@ -1676,9 +1676,9 @@ class Program(object):
         for block_idx in range(len(self.blocks), self.desc.num_blocks()):
             self.blocks.append(Block(self, block_idx))
         for block in self.blocks:
-            block.sync_with_cpp()
+            block._sync_with_cpp()
 
-    def copy_param_info_from(self, other):
+    def _copy_param_info_from(self, other):
         """
         Copy the information of parameters from other program.
 
@@ -1692,13 +1692,13 @@ class Program(object):
             None
         """
         if not isinstance(other, Program):
-            raise TypeError("copy_param_info_from should be invoked with "
+            raise TypeError("_copy_param_info_from should be invoked with "
                             "Program")
 
         if len(self.blocks) != len(other.blocks):
-            raise ValueError("copy_param_info_from should be invoked with two "
+            raise ValueError("_copy_param_info_from should be invoked with two "
                              "program, with represent the same topology")
-        self.global_block().copy_param_info_from(other.global_block())
+        self.global_block()._copy_param_info_from(other.global_block())
 
     def copy_data_info_from(self, other):
         """
@@ -1714,11 +1714,11 @@ class Program(object):
             None
         """
         if not isinstance(other, Program):
-            raise TypeError("copy_param_info_from should be invoked with "
+            raise TypeError("_copy_param_info_from should be invoked with "
                             "Program")
 
         if len(self.blocks) != len(other.blocks):
-            raise ValueError("copy_param_info_from should be invoked with two "
+            raise ValueError("_copy_param_info_from should be invoked with two "
                              "program, with represent the same topology")
         for var in other.global_block().vars.itervalues():
             if var.is_data:

python/paddle/fluid/initializer.py

@@ -148,7 +148,7 @@ class ConstantInitializer(Initializer):
         assert isinstance(var, framework.Variable)
         assert isinstance(block, framework.Block)
         # Initialization Ops should be prepended and not appended
-        op = block.prepend_op(
+        op = block._prepend_op(
             type="fill_constant",
             outputs={"Out": var},
             attrs={
@@ -202,7 +202,7 @@ class UniformInitializer(Initializer):
         # Initialization Ops should be prepended and not appended
         if self._seed == 0:
             self._seed = block.program.random_seed
-        op = block.prepend_op(
+        op = block._prepend_op(
             type="uniform_random",
             outputs={"Out": var},
             attrs={
@@ -256,7 +256,7 @@ class NormalInitializer(Initializer):
         # Initialization Ops should be prepended and not appended
         if self._seed == 0:
             self._seed = block.program.random_seed
-        op = block.prepend_op(
+        op = block._prepend_op(
             type="gaussian_random",
             outputs={"Out": var},
             attrs={
@@ -346,7 +346,7 @@ class XavierInitializer(Initializer):
 
         if self._uniform:
             limit = np.sqrt(6.0 / float(fan_in + fan_out))
-            op = block.prepend_op(
+            op = block._prepend_op(
                 type="uniform_random",
                 outputs={"Out": var},
                 attrs={
@@ -359,7 +359,7 @@ class XavierInitializer(Initializer):
 
         else:
             std = np.sqrt(2.0 / float(fan_in + fan_out))
-            op = block.prepend_op(
+            op = block._prepend_op(
                 type="gaussian_random",
                 outputs={"Out": var},
                 attrs={
@@ -444,7 +444,7 @@ class MSRAInitializer(Initializer):
 
         if self._uniform:
             limit = np.sqrt(6.0 / float(fan_in))
-            op = block.prepend_op(
+            op = block._prepend_op(
                 type="uniform_random",
                 outputs={"Out": var},
                 attrs={
@@ -457,7 +457,7 @@ class MSRAInitializer(Initializer):
 
         else:
             std = np.sqrt(2.0 / float(fan_in))
-            op = block.prepend_op(
+            op = block._prepend_op(
                 type="gaussian_random",
                 outputs={"Out": var},
                 attrs={

python/paddle/fluid/io.py

@@ -523,7 +523,7 @@ def prepend_feed_ops(inference_program,
 
     for i, name in enumerate(feed_target_names):
         out = global_block.var(name)
-        global_block.prepend_op(
+        global_block._prepend_op(
             type='feed',
             inputs={'X': [feed_var]},
             outputs={'Out': [out]},
@@ -625,7 +625,7 @@ def save_inference_model(dirname,
     for i, op in enumerate(global_block.ops):
         op.desc.set_is_target(False)
         if op.type == "feed" or op.type == "fetch":
-            global_block.remove_op(i)
+            global_block._remove_op(i)
     copy_program.desc.flush()
 
     pruned_program = copy_program.prune(targets=target_vars)
@@ -874,7 +874,7 @@ def get_test_program(filelist, program=None, startup_program=None):
     main_block = program.global_block()
     for var in main_block.vars.values():
         if var.type == core.VarDesc.VarType.READER:
-            main_block.rename_var(
+            main_block._rename_var(
                 str(var.name), str(_get_test_reader_name(var.name)))
 
     for op in main_block.ops:
@@ -883,7 +883,7 @@ def get_test_program(filelist, program=None, startup_program=None):
         if op.type == "create_multi_pass_reader":
             test_op.set_attr("pass_num", 1)
 
-    startup_program.sync_with_cpp()
-    program.sync_with_cpp()
+    startup_program._sync_with_cpp()
+    program._sync_with_cpp()
 
     return program

python/paddle/fluid/layers/__init__.py

@@ -33,7 +33,6 @@ from metric_op import *
 from learning_rate_scheduler import *
 
 __all__ = []
-__all__ += math_op_patch.__all__
 __all__ += nn.__all__
 __all__ += io.__all__
 __all__ += tensor.__all__

python/paddle/fluid/layers/control_flow.py

@@ -730,8 +730,10 @@ class While(object):
         parent_block.append_op(
             type='while',
             inputs={
-                'X':
-                [parent_block.var_recursive(x_name) for x_name in x_name_list],
+                'X': [
+                    parent_block._var_recursive(x_name)
+                    for x_name in x_name_list
+                ],
                 'Condition': [self.cond_var]
             },
             outputs={'Out': out_vars,
@@ -1259,7 +1261,7 @@ class ConditionalBlock(object):
         input_set = set([ipt.name for ipt in self.inputs])
 
         param_list = [
-            parent_block.var_recursive(each_name) for each_name in params
+            parent_block._var_recursive(each_name) for each_name in params
             if each_name not in input_set
         ]
 

python/paddle/fluid/layers/math_op_patch.py

@@ -16,8 +16,6 @@ from ..framework import Variable, unique_name
 from layer_function_generator import OpProtoHolder
 from ..initializer import force_init_on_cpu
 
-__all__ = ['monkey_patch_variable']
-
 
 def monkey_patch_variable():
     def unique_tmp_name():

python/paddle/fluid/layers/nn.py

@@ -4367,7 +4367,7 @@ def autoincreased_step_counter(counter_name=None, begin=1, step=1):
         helper.set_variable_initializer(
             counter, initializer=Constant(
                 value=begin - 1, force_cpu=True))
-        helper.main_program.global_block().prepend_op(
+        helper.main_program.global_block()._prepend_op(
             type='increment',
             inputs={'X': [counter]},
             outputs={'Out': [counter]},

python/paddle/fluid/optimizer.py

@@ -240,7 +240,7 @@ class Optimizer(object):
             self._finish_update(loss.block, parameters_and_grads)
 
             end = len(global_block.ops)
-            return global_block.slice_ops(start, end)
+            return global_block._slice_ops(start, end)
 
     def minimize(self,
                  loss,

python/paddle/fluid/parallel_executor.py

@@ -152,7 +152,7 @@ class ParallelExecutor(object):
         self.executor = core.ParallelExecutor(
             self._places,
             set([
-                p.name for p in main.global_block().iter_parameters()
+                p.name for p in main.global_block()._iter_parameters()
                 if not p.stop_gradient
             ]),
             set(self.persistable_vars), main.desc, loss_name

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

@@ -181,13 +181,13 @@ class TestBlockDesc(unittest.TestCase):
         self.assertIsNotNone(block)
         op1 = block.append_op()
         op2 = block.append_op()
-        op0 = block.prepend_op()
+        op0 = block._prepend_op()
         all_ops = []
         for idx in xrange(0, block.op_size()):
             all_ops.append(block.op(idx))
         self.assertEqual(all_ops, [op0, op1, op2])
 
-    def test_remove_op(self):
+    def test__remove_op(self):
         program = Program()
         program_desc = program.desc
         self.assertIsNotNone(program_desc)
@@ -201,8 +201,8 @@ class TestBlockDesc(unittest.TestCase):
         op1.set_type("test")
         op2.set_type("test")
 
-        block.remove_op(1, 2)
-        program.sync_with_cpp()
+        block._remove_op(1, 2)
+        program._sync_with_cpp()
 
         all_ops = []
         for idx in xrange(0, block.op_size()):

python/paddle/fluid/transpiler/details/program_utils.py

@@ -17,10 +17,10 @@ def delete_ops(block, ops):
     try:
         start = list(block.ops).index(ops[0])
         end = list(block.ops).index(ops[-1])
-        [block.remove_op(start) for _ in xrange(end - start + 1)]
+        [block._remove_op(start) for _ in xrange(end - start + 1)]
     except Exception, e:
         raise e
-    block.program.sync_with_cpp()
+    block.program._sync_with_cpp()
 
 
 def find_op_by_input_arg(block, arg_name):

python/paddle/fluid/transpiler/distribute_transpiler.py

@@ -243,7 +243,7 @@ class DistributeTranspiler(object):
                 AssertionError("Can not insert the send op by original "
                                "variable name :", orig_varname)
 
-            program.global_block().insert_op(
+            program.global_block()._insert_op(
                 index=index + 1,
                 type="send",
                 inputs={"X": splited_vars},
@@ -429,7 +429,7 @@ class DistributeTranspiler(object):
 
             # clone vars
             for var in origin_block.vars:
-                new_sub_block.clone_variable(var)
+                new_sub_block._clone_variable(var)
 
             # clone ops
             for origin_op in origin_block.ops:
@@ -525,7 +525,7 @@ class DistributeTranspiler(object):
             outputs={},
             attrs=attrs)
 
-        pserver_program.sync_with_cpp()
+        pserver_program._sync_with_cpp()
         return pserver_program
 
     def get_startup_program(self, endpoint, pserver_program):
@@ -557,7 +557,7 @@ class DistributeTranspiler(object):
         pserver_vars = pserver_program.global_block().vars
         created_var_map = dict()
         for _, var in pserver_vars.iteritems():
-            tmpvar = s_prog.global_block().clone_variable(var)
+            tmpvar = s_prog.global_block()._clone_variable(var)
             created_var_map[var.name] = tmpvar
 
         # 2. rename op outputs
@@ -760,7 +760,7 @@ class DistributeTranspiler(object):
                     self.all_prefetch_output_vars.append(prefetch_output_vars)
 
                     # insert split_ids_op
-                    program.global_block().insert_op(
+                    program.global_block()._insert_op(
                         index=lookup_table_op_index,
                         type="split_ids",
                         inputs={
@@ -772,7 +772,7 @@ class DistributeTranspiler(object):
                         outputs={"Out": prefetch_input_vars})
 
                     # insert prefetch_op
-                    program.global_block().insert_op(
+                    program.global_block()._insert_op(
                         index=lookup_table_op_index + 1,
                         type="prefetch",
                         inputs={'X': prefetch_input_vars},
@@ -783,7 +783,7 @@ class DistributeTranspiler(object):
                         })
 
                     # insert concat_op
-                    program.global_block().insert_op(
+                    program.global_block()._insert_op(
                         index=lookup_table_op_index + 2,
                         type="merge_ids",
                         inputs={
@@ -814,14 +814,14 @@ class DistributeTranspiler(object):
             if table_grad_name in op.output_arg_names:
                 op_index = list(all_ops).index(op)
                 # insert split_ids_op
-                program.global_block().insert_op(
+                program.global_block()._insert_op(
                     index=op_index + 1,
                     type="split_ids",
                     inputs={
                         'Ids': [program.global_block().vars[table_grad_name]]
                     },
                     outputs={"Out": self.trainer_side_table_grad_list})
-                program.global_block().insert_op(
+                program.global_block()._insert_op(
                     index=op_index + 2,
                     type="send",
                     inputs={'X': self.trainer_side_table_grad_list},
@@ -880,7 +880,7 @@ class DistributeTranspiler(object):
             persistable=True)
         # parameter must be selected rows
         param_var.desc.set_type(core.VarDesc.VarType.SELECTED_ROWS)
-        grad_var = pserver_program.global_block().clone_variable(
+        grad_var = pserver_program.global_block()._clone_variable(
             self.origin_program.global_block().vars[grad_var_name(
                 self.table_name)])
 
@@ -920,7 +920,7 @@ class DistributeTranspiler(object):
             if not splited_grad_name.startswith(origin_grad_name):
                 raise ValueError("origin_grad_var: " + splited_grad_name +
                                  " grad_var:" + grad_var.name)
-            grad_var = pserver_program.global_block().rename_var(
+            grad_var = pserver_program.global_block()._rename_var(
                 origin_grad_name, splited_grad_name)
 
         lr_var = pserver_program.global_block().vars[table_opt_op.input(
@@ -996,7 +996,7 @@ class DistributeTranspiler(object):
                 if self.sync_mode and add_trainer_suffix:
                     new_var_name = "%s.trainer_%d" % \
                         (orig_var.name, self.trainer_id)
-                    program.global_block().rename_var(varname, new_var_name)
+                    program.global_block()._rename_var(varname, new_var_name)
                     var_mapping[varname] = \
                         [program.global_block().var(new_var_name)]
                 else:
@@ -1030,8 +1030,7 @@ class DistributeTranspiler(object):
                     type=orig_var.type,
                     shape=splited_shape)  # flattend splited var
                 var_mapping[varname].append(var)
-            program.global_block().sync_with_cpp()
-
+            program.global_block()._sync_with_cpp()
         return var_mapping
 
     def create_splited_vars(self, source_var, block, tag):
@@ -1059,7 +1058,7 @@ class DistributeTranspiler(object):
             height_sections = []
             for v in splited_vars:
                 height_sections.append(v.shape[0])
-            program.global_block().insert_op(
+            program.global_block()._insert_op(
                 index=index + 1,
                 type="split_selected_rows",
                 inputs={"X": orig_var},
@@ -1069,7 +1068,7 @@ class DistributeTranspiler(object):
             sections = []
             for v in splited_vars:
                 sections.append(v.shape[0])
-            program.global_block().insert_op(
+            program.global_block()._insert_op(
                 index=index + 1,
                 type="split_byref",
                 inputs={"X": orig_var},
@@ -1258,7 +1257,7 @@ class DistributeTranspiler(object):
                 varlist = [varlist]
             for var in varlist:
                 if var not in program.global_block().vars:
-                    block.clone_variable(var)
+                    block._clone_variable(var)
 
         outputs = self._get_output_map_from_op(
             self.origin_program.global_block().vars, op)
@@ -1267,7 +1266,7 @@ class DistributeTranspiler(object):
                 varlist = [varlist]
             for var in varlist:
                 if var not in program.global_block().vars:
-                    block.clone_variable(var)
+                    block._clone_variable(var)
 
         return block.append_op(
             type=op.type, inputs=inputs, outputs=outputs, attrs=op.attrs)
@@ -1305,7 +1304,7 @@ class DistributeTranspiler(object):
                 if grad_block:
                     outputs[key] = grad_block
                 elif not program.global_block().vars.has_key(var.name):
-                    program.global_block().clone_variable(var)
+                    program.global_block()._clone_variable(var)
 
         return optimize_block.append_op(
             type=opt_op.type,

python/paddle/fluid/transpiler/inference_transpiler.py

@@ -95,7 +95,7 @@ class InferenceTranspiler(object):
                     # modify bnorm OP to include relu
                     current_op.set_attr("fuse_with_relu", True)
                     # remove relu OP
-                    self.block.remove_op(i + 1)
+                    self.block._remove_op(i + 1)
             i = i + 1
 
         self._remove_unused_var()
@@ -171,7 +171,7 @@ class InferenceTranspiler(object):
                     # fuse batch_norm
                     self._fuse_param(current_op, next_op, bias_op, 0)
                     # remove batch_norm_op
-                    self.block.remove_op(i + 2)
+                    self.block._remove_op(i + 2)
                     i = i + 1
                 # conv2d with bias, the next_op.type is elementwise_add
                 elif (next_op.type == 'elementwise_add'):
@@ -180,7 +180,7 @@ class InferenceTranspiler(object):
                         # fuse batch_norm
                         self._fuse_param(current_op, next_next_op, next_op, 1)
                         # remove batch_norm_op
-                        self.block.remove_op(i + 2)
+                        self.block._remove_op(i + 2)
                         i = i + 1
             i = i + 1
 
@@ -212,7 +212,7 @@ class InferenceTranspiler(object):
         y_var = self.block.var(bn_op.input("Bias")[0])
         out_var = self.block.var(bn_op.output("Y")[0])
 
-        bias_op = self.block.insert_op(
+        bias_op = self.block._insert_op(
             index,
             type="elementwise_add",
             inputs={"X": x_var,
@@ -307,4 +307,4 @@ class InferenceTranspiler(object):
 
         for var in self.block.vars.keys():
             if var not in args:
-                self.block.remove_var(var)
+                self.block._remove_var(var)

python/paddle/fluid/transpiler/memory_optimization_transpiler.py

@@ -177,7 +177,7 @@ class ControlFlowGraph(object):
                 in_diff)
             if can_optimize:
                 index = i + fwd_id + 1 if is_forward else i - self._forward_num + bwd_id + 1
-                delete_op = block_desc.insert_op(index)
+                delete_op = block_desc._insert_op(index)
                 delete_op.set_type("delete_var")
                 delete_op.set_input("X", can_optimize)
                 if is_forward:

python/paddle/v2/image.py

@@ -182,7 +182,7 @@ def resize_short(im, size):
         h_new = size * h / w
     else:
         w_new = size * w / h
-    im = cv2.resize(im, (h_new, w_new), interpolation=cv2.INTER_CUBIC)
+    im = cv2.resize(im, (w_new, h_new), interpolation=cv2.INTER_CUBIC)
     return im
 
 
@@ -324,7 +324,6 @@ def simple_transform(im,
         if np.random.randint(2) == 0:
             im = left_right_flip(im, is_color)
     else:
-        im = center_crop(im, crop_size, is_color)
         im = center_crop(im, crop_size, is_color=is_color)
     if len(im.shape) == 3:
         im = to_chw(im)