Add PyDataProvider2 DataConverter to swig api.

* fix recommendation prediction also.

ISSUE=4561941



git-svn-id: https://svn.baidu.com/idl/trunk/paddle@1449 1ad973e4-5ce8-4261-8a94-b56d1f490c56

demo/recommendation/.gitignore

@@ -7,3 +7,4 @@ data/train.list
 data/test.list
 dataprovider_copy_1.py
 *.pyc
+output

demo/recommendation/prediction.py

@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from py_paddle import swig_paddle, DataProviderWrapperConverter
+from py_paddle import swig_paddle, DataProviderConverter
 
 from common_utils import *
 from paddle.trainer.config_parser import parse_config
@@ -31,11 +31,11 @@
     network = swig_paddle.GradientMachine.createFromConfigProto(conf.model_config)
     assert isinstance(network, swig_paddle.GradientMachine)
     network.loadParameters(model_path)
-    with open('meta.bin', 'rb') as f:
+    with open('./data/meta.bin', 'rb') as f:
         meta = pickle.load(f)
         headers = list(meta_to_header(meta, 'movie'))
         headers.extend(list(meta_to_header(meta, 'user')))
-        cvt = DataProviderWrapperConverter(True, map(lambda x: x[1], headers))
+        cvt = DataProviderConverter(headers)
         while True:
             movie_id = int(raw_input("Input movie_id: "))
             user_id = int(raw_input("Input user_id: "))
@@ -45,7 +45,5 @@
             data.extend(movie_meta)
             data.append(user_id - 1)
             data.extend(user_meta)
-            data = map(lambda (header, val): val if header[0] else [val],
-                       zip(headers, data))
-            print "Prediction Score is %.2f" % ((network.forwardTest(cvt([
-                data]))[0]['value'][0][0] + 5) / 2)
+            print "Prediction Score is %.2f" % ((network.forwardTest(
+                cvt.convert([data]))[0]['value'][0][0] + 5) / 2)

demo/recommendation/trainer_config.py

@@ -61,9 +61,7 @@ def construct_feature(name):
             slot_dim = each_meta['max']
             embedding = embedding_layer(input=data_layer(slot_name,
                                                           size=slot_dim),
-                                        size=256,
-                                        param_attr=ParamAttr(
-                                            sparse_update=True))
+                                        size=256)
             fusion.append(fc_layer(input=embedding,
                                    size=256))
         elif type_name == 'embedding':

paddle/api/Arguments.cpp

@@ -102,8 +102,23 @@ static inline void doCopyFromSafely(std::shared_ptr<T1>& dest,
 IVector* Arguments::getSlotSequenceStartPositions(size_t idx) const
     throw(RangeError) {
   auto& a = m->getArg(idx);
-  return IVector::createByPaddleVectorPtr(
-    &a.sequenceStartPositions->getMutableVector(false));
+  if (a.sequenceStartPositions) {
+    return IVector::createByPaddleVectorPtr(
+        &a.sequenceStartPositions->getMutableVector(false));
+  } else {
+    return nullptr;
+  }
+}
+
+IVector*Arguments::getSlotSubSequenceStartPositions(size_t idx) const
+    throw (RangeError){
+  auto& a = m->getArg(idx);
+  if (a.subSequenceStartPositions) {
+    return IVector::createByPaddleVectorPtr(
+        &a.subSequenceStartPositions->getMutableVector(false));
+  } else {
+    return nullptr;
+  }
 }
 
 void Arguments::setSlotSequenceStartPositions(size_t idx,
@@ -113,6 +128,13 @@ void Arguments::setSlotSequenceStartPositions(size_t idx,
   a.sequenceStartPositions = std::make_shared<paddle::ICpuGpuVector>(v);
 }
 
+void Arguments::setSlotSubSequenceStartPositions(
+    size_t idx, IVector *vec) throw (RangeError) {
+  auto& a = m->getArg(idx);
+  auto& v = m->cast<paddle::IVector>(vec->getSharedPtr());
+  a.subSequenceStartPositions = std::make_shared<paddle::ICpuGpuVector>(v);
+}
+
 IVector* Arguments::getSlotSequenceDim(size_t idx) const throw(RangeError) {
   auto& a = m->getArg(idx);
   return IVector::createByPaddleVectorPtr(&a.cpuSequenceDims);

paddle/api/PaddleAPI.h

@@ -374,6 +374,7 @@ class Arguments {
   IVector* getSlotIds(size_t idx) const throw(RangeError);
   Matrix* getSlotIn(size_t idx) const throw(RangeError);
   IVector* getSlotSequenceStartPositions(size_t idx) const throw(RangeError);
+  IVector* getSlotSubSequenceStartPositions(size_t idx) const throw(RangeError);
   IVector* getSlotSequenceDim(size_t idx) const throw(RangeError);
   // End Of get functions of Arguments
 
@@ -390,6 +391,8 @@ class Arguments {
   void setSlotIds(size_t idx, IVector* vec) throw(RangeError);
   void setSlotSequenceStartPositions(size_t idx,
                                      IVector* vec) throw(RangeError);
+  void setSlotSubSequenceStartPositions(size_t idx,
+                                        IVector* vec) throw (RangeError);
   void setSlotSequenceDim(size_t idx, IVector* vec) throw(RangeError);
 
 private:

paddle/py_paddle/__init__.py

@@ -12,10 +12,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import swig_paddle
-import util
 from util import DataProviderWrapperConverter
+from dataprovider_converter import DataProviderConverter
 
-__all__ = ['paddle', 'DataProviderWrapperConverter', 'loadParameterFile']
-
+__all__ = ['paddle',
+           'DataProviderConverter',
+           'DataProviderWrapperConverter',  # for deprecated usage.
+           'loadParameterFile']
 util.monkeypatches()
+

paddle/py_paddle/util.py

@@ -209,6 +209,7 @@ def getLayerOutputs(self, layerNames):
 
     swig_paddle.GradientMachine.getLayerOutputs = getLayerOutputs
 
+
 def loadGradientMachine(config_filename, model_dir=None):
     """
     Load a gradient machine from config file name/path.
@@ -229,6 +230,7 @@ def loadGradientMachine(config_filename, model_dir=None):
     network.loadParameters(model_dir)
     return network
 
+
 def loadParameterFile(fn):
     """
     Load Paddle Parameter file to numpy.ndarray
@@ -257,6 +259,7 @@ def loadParameterFile(fn):
     else:
         raise swig_paddle.UnsupportError()
 
+
 class DataProviderWrapperConverter(object):
     """
     A class convert DataFormat from PyDataProvider Wrapper to
@@ -278,8 +281,8 @@ def append(self, other):
 
         def __call__(self, slot_idx, arg):
             mat = swig_paddle.Matrix.createDense(self.buf,
-                                            len(self.buf) / self.__dim__,
-                                            self.__dim__)
+                                                 len(self.buf) / self.__dim__,
+                                                 self.__dim__)
             arg.setSlotValue(slot_idx, mat)
 
     class IdValueConverter(object):
@@ -312,8 +315,9 @@ def append(self, other):
             self.cols += other
 
         def __call__(self, slot_idx, arg):
-            mat = swig_paddle.Matrix.createSparse(len(self.indices) - 1, self.dim,
-                                             len(self.cols), True)
+            mat = swig_paddle.Matrix.createSparse(len(self.indices) - 1,
+                                                  self.dim,
+                                                  len(self.cols), True)
             assert isinstance(mat, swig_paddle.Matrix)
             mat.sparseCopyFrom(self.indices, self.cols)
             self.putIntoArg(slot_idx, arg, mat)
@@ -337,8 +341,9 @@ def append(self, other):
             self.values += map(lambda x: x[1], other)
 
         def __call__(self, slot_idx, arg):
-            mat = swig_paddle.Matrix.createSparse(len(self.indices) - 1, self.dim,
-                                             len(self.cols), False)
+            mat = swig_paddle.Matrix.createSparse(len(self.indices) - 1,
+                                                  self.dim,
+                                                  len(self.cols), False)
             assert isinstance(mat, swig_paddle.Matrix)
             mat.sparseCopyFrom(self.indices, self.cols, self.values)
             self.putIntoArg(slot_idx, arg, mat)
@@ -373,7 +378,7 @@ def convert(self, wrapper_data, argument=None):
         """
         if argument is None:
             argument = swig_paddle.Arguments.createArguments(0)
-        assert isinstance(argument,swig_paddle.Arguments)
+        assert isinstance(argument, swig_paddle.Arguments)
         argument.resize(len(self.__header__))
 
         values = map(lambda x:
@@ -394,10 +399,12 @@ def convert(self, wrapper_data, argument=None):
                     seq_dim[slot_idx].append(len(sequence))
 
             for slot_idx in xrange(len(self.__header__)):
-                argument.setSlotSequenceDim(slot_idx, swig_paddle.IVector.create(
-                    seq_dim[slot_idx]))
+                argument.setSlotSequenceDim(slot_idx,
+                                            swig_paddle.IVector.create(
+                                                seq_dim[slot_idx]))
                 argument.setSlotSequenceStartPositions(
-                    slot_idx, swig_paddle.IVector.create(seq_start_pos[slot_idx]))
+                    slot_idx,
+                    swig_paddle.IVector.create(seq_start_pos[slot_idx]))
         else:
             for each_sample in wrapper_data:
                 for raw_data, value in zip(each_sample, values):
@@ -415,6 +422,7 @@ def __call__(self, wrapper_data, argument=None):
         return self.convert(wrapper_data, argument)
 
 
+
 def __monkey_patch_protobuf_objects__():
     def ParameterConfig_toProto(self):
         """
@@ -451,7 +459,8 @@ def OptimizationConfig_createFromProto(protoObj):
         :return: paddle.OptimizationConfig
         """
 
-        assert isinstance(protoObj, paddle.proto.TrainerConfig_pb2.OptimizationConfig)
+        assert isinstance(protoObj,
+                          paddle.proto.TrainerConfig_pb2.OptimizationConfig)
         return swig_paddle.OptimizationConfig.createFromProtoString(
             protoObj.SerializeToString())