Get OpProtos in Python

paddle/framework/op_registry.h

@@ -125,17 +125,17 @@ class OpRegistry {
     return op;
   }
 
+  static std::unordered_map<std::string, OpProto>& protos() {
+    static std::unordered_map<std::string, OpProto> protos_;
+    return protos_;
+  };
+
  private:
   static std::unordered_map<std::string, OpCreator>& creators() {
     static std::unordered_map<std::string, OpCreator> creators_;
     return creators_;
   }
 
-  static std::unordered_map<std::string, OpProto>& protos() {
-    static std::unordered_map<std::string, OpProto> protos_;
-    return protos_;
-  };
-
   static std::unordered_map<std::string, OpAttrChecker>& op_checkers() {
     static std::unordered_map<std::string, OpAttrChecker> op_checkers_;
     return op_checkers_;

paddle/pybind/CMakeLists.txt

@@ -1 +1 @@
-cc_library(paddle_pybind SHARED SRCS pybind.cc DEPS pybind python)
+cc_library(paddle_pybind SHARED SRCS pybind.cc DEPS pybind python add_op)

paddle/pybind/pybind.cc

@@ -13,12 +13,18 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include <Python.h>
+#include <paddle/framework/op_registry.h>
 #include <paddle/framework/scope.h>
 #include <pybind11/pybind11.h>
+#include <pybind11/stl.h>
+#include <fstream>
+#include <vector>
 
 namespace py = pybind11;
 namespace pd = paddle::framework;
 
+USE_OP(add_two);
+
 PYBIND11_PLUGIN(core) {
   py::module m("core", "C++ core of Paddle Paddle");
 
@@ -43,5 +49,20 @@ All parameter, weight, gradient are variables in Paddle.
            &pd::Scope::CreateVariable,
            py::return_value_policy::reference);
 
+  //! @note: Be careful! PyBind will return std::string as an unicode, not
+  //! Python str. If you want a str object, you should cast them in Python.
+  m.def("get_all_op_protos", []() -> std::vector<std::string> {
+    auto& protos = pd::OpRegistry::protos();
+    std::vector<std::string> ret_values;
+    for (auto it = protos.begin(); it != protos.end(); ++it) {
+      PADDLE_ENFORCE(it->second.IsInitialized(),
+                     "OpProto must all be initialized");
+      ret_values.emplace_back();
+      PADDLE_ENFORCE(it->second.SerializeToString(&ret_values.back()),
+                     "Serialize OpProto Error. This could be a bug of Paddle.");
+    }
+    return ret_values;
+  });
+
   return m.ptr();
 }

python/paddle/v2/__init__.py

@@ -20,7 +20,6 @@
 import event
 import data_type
 import topology
-import data_feeder
 import networks
 import evaluator
 from . import dataset
@@ -31,7 +30,6 @@
 import pooling
 import inference
 import networks
-import py_paddle.swig_paddle as api
 import minibatch
 import plot
 import image
@@ -47,7 +45,6 @@
     'data_type',
     'attr',
     'pooling',
-    'data_feeder',
     'dataset',
     'reader',
     'topology',
@@ -61,6 +58,7 @@
 
 
 def init(**kwargs):
+    import py_paddle.swig_paddle as api
     args = []
     args_dict = {}
     # NOTE: append arguments if they are in ENV

python/paddle/v2/data_feeder.py

@@ -11,7 +11,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 from py_paddle import DataProviderConverter
 import collections
 import paddle.trainer.PyDataProvider2 as pydp2

python/paddle/v2/event.py

@@ -9,15 +9,14 @@
 * BeginPass
 * EndPass
 """
-import py_paddle.swig_paddle as api
-
 __all__ = [
     'EndIteration', 'BeginIteration', 'BeginPass', 'EndPass', 'TestResult'
 ]
 
 
 class WithMetric(object):
     def __init__(self, evaluator):
+        import py_paddle.swig_paddle as api
         if not isinstance(evaluator, api.Evaluator):
             raise TypeError("Evaluator should be api.Evaluator type")
         self.__evaluator__ = evaluator

python/paddle/v2/framework/create_op_creation_methods.py

@@ -0,0 +1,11 @@
+import paddle.v2.framework.core as core
+import paddle.v2.framework.proto.op_proto_pb2 as op_proto_pb2
+
+
+def get_all_op_protos():
+    protostrs = core.get_all_op_protos()
+    ret_values = []
+    for pbstr in protostrs:
+        op_proto = op_proto_pb2.OpProto.FromString(str(pbstr))
+        ret_values.append(op_proto)
+    return ret_values

python/paddle/v2/framework/tests/CMakeLists.txt

@@ -1,2 +1,2 @@
 add_python_test(test_framework test_protobuf.py test_scope.py
-    test_default_scope_funcs.py)
+    test_default_scope_funcs.py test_op_creation_methods.py)

python/paddle/v2/framework/tests/test_op_creation_methods.py

@@ -0,0 +1,15 @@
+import unittest
+import paddle.v2.framework.create_op_creation_methods as creation
+
+
+class TestOpCreationsMethods(unittest.TestCase):
+    def test_all_protos(self):
+        all_protos = creation.get_all_op_protos()
+        self.assertNotEqual(0, len(all_protos))
+
+        for each in all_protos:
+            self.assertTrue(each.IsInitialized())
+
+
+if __name__ == "__main__":
+    unittest.main()

python/paddle/v2/inference.py

@@ -1,9 +1,7 @@
 import numpy
-import py_paddle.swig_paddle as api
 import collections
 import topology
 import minibatch
-from data_feeder import DataFeeder
 
 __all__ = ['infer', 'Inference']
 
@@ -28,6 +26,7 @@ class Inference(object):
     """
 
     def __init__(self, output_layer, parameters):
+        import py_paddle.swig_paddle as api
         topo = topology.Topology(output_layer)
         gm = api.GradientMachine.createFromConfigProto(
             topo.proto(), api.CREATE_MODE_TESTING, [api.PARAMETER_VALUE])
@@ -40,6 +39,7 @@ def __init__(self, output_layer, parameters):
         self.__data_types__ = topo.data_type()
 
     def iter_infer(self, input, feeding=None):
+        from data_feeder import DataFeeder
         feeder = DataFeeder(self.__data_types__, feeding)
         batch_size = len(input)
 

python/paddle/v2/optimizer.py

@@ -1,5 +1,3 @@
-import py_paddle.swig_paddle as swig_api
-
 import paddle.trainer_config_helpers.config_parser_utils as config_parser_utils
 import paddle.trainer_config_helpers.optimizers as v1_optimizers
 """
@@ -18,6 +16,7 @@
 
 class Optimizer(object):
     def __init__(self, **kwargs):
+        import py_paddle.swig_paddle as swig_api
         if 'batch_size' in kwargs:
             del kwargs['batch_size']  # not important for python library.
 
@@ -268,6 +267,7 @@ def __init__(self, rho=0.95, epsilon=1e-6, **kwargs):
 L2Regularization = v1_optimizers.L2Regularization
 
 if __name__ == '__main__':
+    import py_paddle.swig_paddle as swig_api
     swig_api.initPaddle('--use_gpu=false')
     for opt in [
             Momentum(), Adam(), Adamax(), AdaGrad(), DecayedAdaGrad(),

python/paddle/v2/parameters.py

@@ -1,5 +1,4 @@
 import numpy as np
-import py_paddle.swig_paddle as api
 from paddle.proto.ParameterConfig_pb2 import ParameterConfig
 import paddle.trainer.config_parser as cp
 import struct
@@ -124,6 +123,7 @@ def __getitem__(self, key):
         :return: parameter value
         :rtype: np.ndarray
         """
+        import py_paddle.swig_paddle as api
         shape = self.get_shape(key)
 
         if len(self.__gradient_machines__) == 0:
@@ -223,7 +223,7 @@ def append_gradient_machine(self, gradient_machine):
         :type gradient_machine: api.GradientMachine
         :return:
         """
-
+        import py_paddle.swig_paddle as api
         if not isinstance(gradient_machine, api.GradientMachine):
             raise ValueError("gradient_machine should be api.GradientMachine")
 
@@ -359,6 +359,7 @@ def __copy_parameter_to_gradient_machine__(gradient_machine, name, arr):
     :return:
     :rtype: api.Parameter
     """
+    import py_paddle.swig_paddle as api
     param = __get_parameter_in_gradient_machine__(gradient_machine, name)
     vec = param.getBuf(api.PARAMETER_VALUE)
     assert isinstance(vec, api.Vector)

python/paddle/v2/trainer.py

@@ -2,12 +2,6 @@
 Module Trainer
 """
 import collections
-import gzip
-import os
-
-import py_paddle.swig_paddle as api
-
-from data_feeder import DataFeeder
 from topology import Topology
 from . import event as v2_event
 from . import optimizer as v2_optimizer
@@ -59,6 +53,7 @@ def __init__(self,
         if not isinstance(update_equation, v2_optimizer.Optimizer):
             raise TypeError("update equation parameter must be "
                             "paddle.v2.optimizer.Optimizer")
+        import py_paddle.swig_paddle as api
         topology = Topology(cost, extra_layers=extra_layers)
         self.__optimizer__ = update_equation
         self.__topology__ = topology
@@ -124,6 +119,8 @@ def train(self, reader, num_passes=1, event_handler=None, feeding=None):
         :type feeding: dict|list
         :return:
         """
+        import py_paddle.swig_paddle as api
+        from data_feeder import DataFeeder
         if event_handler is None:
             event_handler = default_event_handler
         __check_train_args__(**locals())
@@ -187,6 +184,8 @@ def test(self, reader, feeding=None):
         :type feeding: dict
         :return:
         """
+        import py_paddle.swig_paddle as api
+        from data_feeder import DataFeeder
         feeder = DataFeeder(self.__data_types__, feeding)
         evaluator = self.__gradient_machine__.makeEvaluator()
         out_args = api.Arguments.createArguments(0)

python/setup.py.in

@@ -19,7 +19,8 @@ setup_requires=["requests",
                 "recordio",
                 "matplotlib",
                 "rarfile",
-                "scipy>=0.19.0"]
+                "scipy>=0.19.0",
+                "nltk"]
 
 if '${CMAKE_SYSTEM_PROCESSOR}' not in ['arm', 'armv7-a', 'aarch64']:
     setup_requires+=["opencv-python"]