diff --git a/python/paddle/fluid/__init__.py b/python/paddle/fluid/__init__.py
index 3676ffe938e4c174547f0c70d94ec35f9156cb1e..e2b49a31d15a2e493be3c3e68baaf602a839fc42 100644
--- a/python/paddle/fluid/__init__.py
+++ b/python/paddle/fluid/__init__.py
@@ -72,7 +72,11 @@ Tensor = LoDTensor
 __all__ = framework.__all__ + executor.__all__ + \
     trainer_desc.__all__ + inferencer.__all__ + transpiler.__all__ + \
     parallel_executor.__all__ + lod_tensor.__all__ + \
+<<<<<<< HEAD
     data_feed_desc.__all__ + async_executor.__all__ + compiler.__all__  + [
+=======
+    data_feed_desc.__all__ + async_executor.__all__ + compiler.__all__ + [
+>>>>>>> add data_generator package into setup.py
         'io',
         'initializer',
         'layers',
diff --git a/python/paddle/fluid/executor.py b/python/paddle/fluid/executor.py
index a14f4e2700b10a65e5a869057ac611268711d1d9..d609b88fe5e755c8e2d6a065843324898978df38 100644
--- a/python/paddle/fluid/executor.py
+++ b/python/paddle/fluid/executor.py
@@ -654,7 +654,7 @@ class Executor(object):
             trainer._set_thread(thread)
         trainer._set_debug(debug)
         trainer._set_fetch_var_and_info(fetch_list, fetch_info, print_period)
-        return trainer
+        return scope, trainer
 
     def infer_from_dataset(self,
                            program=None,
@@ -702,7 +702,7 @@ class Executor(object):
                                        dataset=dataset)        
         """
 
-        trainer = self._prepare_trainer(
+        scope, trainer = self._prepare_trainer(
             program=program,
             dataset=dataset,
             scope=scope,
@@ -775,7 +775,7 @@ class Executor(object):
 
         """
 
-        trainer = self._prepare_trainer(
+        scope, trainer = self._prepare_trainer(
             program=program,
             dataset=dataset,
             scope=scope,
diff --git a/python/paddle/fluid/trainer_desc.py b/python/paddle/fluid/trainer_desc.py
index 84fdb57839e4c845e8edce5a007afc26c951a6b1..c31ebbd151a5e072a9d33ab34ec5ff6cd3cff215 100644
--- a/python/paddle/fluid/trainer_desc.py
+++ b/python/paddle/fluid/trainer_desc.py
@@ -75,14 +75,14 @@ class MultiTrainer(TrainerDesc):
         pass
 
     def _set_program(self, program):
-        super(MultiTrainer, self).set_program(program)
+        super(MultiTrainer, self)._set_program(program)
         self.program_ = program
 
     def _gen_trainer_desc(self):
-        super(MultiTrainer, self).gen_trainer_desc()
+        super(MultiTrainer, self)._gen_trainer_desc()
         self.proto_desc.class_name = "MultiTrainer"
-        self.device_worker_.set_infer(self.infer_)
-        self.device_worker_.gen_worker_desc(self.proto_desc)
+        self.device_worker_._set_infer(self.infer_)
+        self.device_worker_._gen_worker_desc(self.proto_desc)
 
 
 class DistMultiTrainer(TrainerDesc):
@@ -91,14 +91,14 @@ class DistMultiTrainer(TrainerDesc):
         pass
 
     def _set_program(self, program):
-        super(DistMultiTrainer, self).set_program(program)
+        super(DistMultiTrainer, self)._set_program(program)
         self.program_ = program
 
     def _gen_trainer_desc(self):
-        super(DistMultiTrainer, self).gen_trainer_desc()
+        super(DistMultiTrainer, self)._gen_trainer_desc()
         self.proto_desc.class_name = "DistMultiTrainer"
         if self.program_ == None:
             print("None program")
-        self.device_worker_.set_infer(self.infer_)
-        self.device_worker_.set_program(self.program_)
-        self.device_worker_.gen_worker_desc(self.proto_desc)
+        self.device_worker_._set_infer(self.infer_)
+        self.device_worker_._set_program(self.program_)
+        self.device_worker_._gen_worker_desc(self.proto_desc)
diff --git a/python/paddle/fluid/trainer_factory.py b/python/paddle/fluid/trainer_factory.py
index 8faab282773aad24982453352bca2ff63a0046a6..871b663663e87a08ef3edaf58a4480b85caf4c4a 100644
--- a/python/paddle/fluid/trainer_factory.py
+++ b/python/paddle/fluid/trainer_factory.py
@@ -29,13 +29,13 @@ class TrainerFactory(object):
             # default is MultiTrainer + Hogwild
             trainer = MultiTrainer()
             device_worker = Hogwild()
-            trainer.set_device_worker(device_worker)
+            trainer._set_device_worker(device_worker)
         else:
             trainer_class = opt_info["trainer"]
             device_worker_class = opt_info["device_worker"]
             trainer = globals()[trainer_class]()
             device_worker = globals()[device_worker_class]()
-            device_worker.set_fleet_desc(opt_info["fleet_desc"])
-            trainer.set_device_worker(device_worker)
-            trainer.set_fleet_desc(opt_info["fleet_desc"])
+            device_worker._set_fleet_desc(opt_info["fleet_desc"])
+            trainer._set_device_worker(device_worker)
+            trainer._set_fleet_desc(opt_info["fleet_desc"])
         return trainer
diff --git a/python/setup.py.in b/python/setup.py.in
index 801eef741efac9f06a83f8317156aa5227c9dce4..0630b44b8468a224bb3dc4ac559bf5a2d4be3c31 100644
--- a/python/setup.py.in
+++ b/python/setup.py.in
@@ -122,6 +122,7 @@ packages=['paddle',
           'paddle.fluid.transpiler',
           'paddle.fluid.transpiler.details',
           'paddle.fluid.incubate',
+          'paddle.fluid.incubate.data_generator',
           'paddle.fluid.incubate.fleet',
           'paddle.fluid.incubate.fleet.base',
           'paddle.fluid.incubate.fleet.parameter_server',