Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into topology

demo/image_classification/prediction.py

@@ -126,7 +126,7 @@ class ImageClassifier():
         # For oversampling, average predictions across crops.
         # If not, the shape of output[name]: (1, class_number),
         # the mean is also applicable.
-        return output[output_layer].mean(0)
+        return output[output_layer]['value'].mean(0)
 
     def predict(self, image=None, output_layer=None):
         assert isinstance(image, basestring)

demo/mnist/api_train_v2.py

@@ -1,4 +1,3 @@
-import numpy
 import paddle.v2 as paddle
 
 import mnist_util
@@ -27,19 +26,14 @@ def main():
     cost = paddle.layer.classification_cost(input=inference, label=label)
 
     parameters = paddle.parameters.create(cost)
-    for param_name in parameters.keys():
-        array = parameters.get(param_name)
-        array[:] = numpy.random.uniform(low=-1.0, high=1.0, size=array.shape)
-        parameters.set(parameter_name=param_name, value=array)
 
     adam_optimizer = paddle.optimizer.Adam(learning_rate=0.01)
 
     def event_handler(event):
         if isinstance(event, paddle.event.EndIteration):
-            para = parameters.get('___fc_2__.w0')
-            print "Pass %d, Batch %d, Cost %f, Weight Mean Of Fc 2 is %f" % (
-                event.pass_id, event.batch_id, event.cost, para.mean())
-
+            if event.batch_id % 100 == 0:
+                print "Pass %d, Batch %d, Cost %f, %s" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics)
         else:
             pass
 

demo/model_zoo/resnet/classify.py

@@ -156,7 +156,7 @@ class ImageClassifier():
             # For oversampling, average predictions across crops.
             # If not, the shape of output[name]: (1, class_number),
             # the mean is also applicable.
-            res[name] = output[name].mean(0)
+            res[name] = output[name]['value'].mean(0)
 
         return res
 

doc/api/trainer_config_helpers/layers.rst

@@ -139,24 +139,12 @@ lstmemory
     :members: lstmemory
     :noindex:
 
-lstm_step_layer
----------------
-..  automodule:: paddle.trainer_config_helpers.layers
-    :members: lstm_step_layer
-    :noindex:
-
 grumemory
 ---------
 ..  automodule:: paddle.trainer_config_helpers.layers
     :members: grumemory
     :noindex:
 
-gru_step_layer
----------------
-..  automodule:: paddle.trainer_config_helpers.layers
-    :members: gru_step_layer
-    :noindex:
-
 Recurrent Layer Group
 =====================
 
@@ -172,6 +160,18 @@ recurrent_group
     :members: recurrent_group
     :noindex:
     
+lstm_step_layer
+---------------
+..  automodule:: paddle.trainer_config_helpers.layers
+    :members: lstm_step_layer
+    :noindex:
+
+gru_step_layer
+---------------
+..  automodule:: paddle.trainer_config_helpers.layers
+    :members: gru_step_layer
+    :noindex:
+
 beam_search
 ------------
 ..  automodule:: paddle.trainer_config_helpers.layers
@@ -308,6 +308,12 @@ repeat_layer
     :members: repeat_layer
     :noindex:
 
+rotate_layer
+------------
+..  automodule:: paddle.trainer_config_helpers.layers
+    :members: rotate_layer
+    :noindex:
+
 seq_reshape_layer
 -----------------
 ..  automodule:: paddle.trainer_config_helpers.layers
@@ -462,6 +468,12 @@ ctc_layer
     :members: ctc_layer
     :noindex:
 
+warp_ctc_layer
+--------------
+..  automodule:: paddle.trainer_config_helpers.layers
+    :members: warp_ctc_layer
+    :noindex:
+
 nce_layer
 -----------
 ..  automodule:: paddle.trainer_config_helpers.layers

paddle/api/Arguments.cpp

@@ -38,6 +38,13 @@ Arguments* Arguments::createByPaddleArgumentVector(void* ptr) {
   return args;
 }
 
+Arguments* Arguments::createByPaddleArgument(const void* ptr) {
+  auto p = (paddle::Argument*)(ptr);
+  auto args = new Arguments();
+  args->m->outputs.push_back(*p);
+  return args;
+}
+
 Matrix* Arguments::getSlotValue(size_t idx) const throw(RangeError) {
   auto& a = m->getArg(idx);
   return Matrix::createByPaddleMatrixPtr(&a.value);

paddle/api/Evaluator.cpp

@@ -27,3 +27,18 @@ std::string Evaluator::toString() {
   m->rawPtr->printStats(sout);
   return sout.str();
 }
+
+std::vector<std::string> Evaluator::getNames() const {
+  std::vector<std::string> retv;
+  m->rawPtr->getNames(&retv);
+  return retv;
+}
+
+double Evaluator::getValue(const std::string name) const {
+  paddle::Error err;
+  double v = m->rawPtr->getValue(name, &err);
+  if (err) {
+    throw std::runtime_error(err.msg());
+  }
+  return v;
+}

paddle/api/GradientMachine.cpp

@@ -144,12 +144,12 @@ Parameter* GradientMachine::getParameter(size_t i) throw(RangeError) {
 
 void GradientMachine::randParameters() { m->machine->randParameters(); }
 
-Matrix* GradientMachine::getLayerOutput(const std::string& layerName) const
+Arguments* GradientMachine::getLayerOutput(const std::string& layerName) const
     throw(UnsupportError) {
-  auto nn = std::dynamic_pointer_cast<paddle::NeuralNetwork>(m->machine);
+  auto nn = m->machine;
   if (nn) {
-    auto mat = nn->getLayerOutput(layerName);
-    return Matrix::createByPaddleMatrixPtr(&mat);
+    auto arg = nn->getLayerOutput(layerName);
+    return Arguments::createByPaddleArgument(&arg);
   } else {
     throw UnsupportError();
   }

paddle/api/PaddleAPI.h

@@ -454,6 +454,7 @@ public:
 
 private:
   static Arguments* createByPaddleArgumentVector(void* ptr);
+  static Arguments* createByPaddleArgument(const void* ptr);
   void* getInternalArgumentsPtr() const;
 
 private:
@@ -769,7 +770,7 @@ public:
 
   void randParameters();
 
-  Matrix* getLayerOutput(const std::string& layerName) const
+  Arguments* getLayerOutput(const std::string& layerName) const
       throw(UnsupportError);
 
   /**
@@ -900,6 +901,10 @@ public:
    */
   std::string toString();
 
+  std::vector<std::string> getNames() const;
+
+  double getValue(const std::string name) const;
+
 private:
   EvaluatorPrivate* m;
 
@@ -952,7 +957,7 @@ public:
 
   Arguments* getForwardOutput();
 
-  Matrix* getLayerOutput(const std::string& layerName);
+  Arguments* getLayerOutput(const std::string& layerName) const;
 };
 
 /// the N-Best results generated from one input sequence.

paddle/api/Trainer.cpp

@@ -131,12 +131,11 @@ void Trainer::testOneDataBatch(size_t batchSize, const Arguments& args) {
 void TrainerPrivate::finishTestPeriod() { tester_->finishTestPeriod(); }
 void Trainer::finishTestPeriod() { m->finishTestPeriod(); }
 
-Matrix* Trainer::getLayerOutput(const std::string& layerName) {
-  auto nn = std::dynamic_pointer_cast<paddle::NeuralNetwork>(
-      this->m->getGradientMachine());
+Arguments* Trainer::getLayerOutput(const std::string& layerName) const {
+  auto nn = this->m->getGradientMachine();
   CHECK(nn) << "trainerInternal_.getGradientMachine() is not NeuralNetwork";
-  auto m = nn->getLayerOutput(layerName);
-  return Matrix::createByPaddleMatrixPtr(&m);
+  auto arg = nn->getLayerOutput(layerName);
+  return Arguments::createByPaddleArgument(&arg);
 }
 
 void Trainer::forwardOneBatch(size_t batchSize) {

paddle/api/test/testTrain.py

@@ -89,9 +89,14 @@ def main():
             except Exception as e:
                 print e
 
+        ev = m.makeEvaluator()
+        ev.start()
         m.forwardBackward(inArgs, outArgs, swig_paddle.PASS_TRAIN,
                           update_callback)
-
+        m.eval(ev)
+        ev.finish()
+        for name in ev.getNames():
+            print name, ev.getValue(name)
         for optimizer in optimizers:
             optimizer.finishBatch()
 

paddle/gserver/gradientmachines/GradientMachine.h

@@ -134,6 +134,10 @@ public:
     backward(callback);
   }
 
+  virtual Argument getLayerOutput(const std::string& layerName) {
+    return *((Argument*)nullptr);
+  }
+
   // see comment in Layer.h for the function with the same name
   virtual void resetState() {}
 

paddle/gserver/gradientmachines/MultiGradientMachine.cpp

@@ -282,6 +282,18 @@ void MultiGradientMachine::forwardBackward(const std::vector<Argument>& inArgs,
   backwardImp(callback);
 }
 
+Argument MultiGradientMachine::getLayerOutput(const std::string& layerName) {
+  std::vector<Argument> args;
+  args.reserve(threads_.size());
+
+  for (auto& thread : threads_) {
+    args.push_back(thread->getGradientMachine()->getLayerOutput(layerName));
+  }
+  outLayerArgs_.concat(args, false /* use_gpu */, outArgStream_, passType_);
+
+  return outLayerArgs_;
+}
+
 void MultiGradientMachine::backwardImp(const UpdateCallback& callback) {
   for (size_t i = 0; i < parameters_.size(); i++) {
     if (!parameters_[i]->useGpu() || parameters_[i]->isStatic()) continue;

paddle/gserver/gradientmachines/MultiGradientMachine.h

@@ -189,6 +189,8 @@ public:
                        PassType passType,
                        const UpdateCallback& callback);
 
+  virtual Argument getLayerOutput(const std::string& layerName);
+
   virtual void onPassEnd();
 
   virtual void finish();
@@ -314,6 +316,8 @@ protected:
   std::vector<Argument> outArgs_;
   hl_stream_t outArgStream_;
 
+  Argument outLayerArgs_;
+
   /// ParameterType which needs to be merged from each GPU
   std::vector<ParameterType> mergeTypes_;
   int numDevices_;         /* number of gpu devices */

paddle/gserver/gradientmachines/NeuralNetwork.cpp

@@ -293,11 +293,10 @@ void NeuralNetwork::backward(const UpdateCallback& callback) {
   }
 }
 
-MatrixPtr NeuralNetwork::getLayerOutput(const std::string& layerName) {
-  auto it = layerMap_.find(layerName);
-  CHECK(it != layerMap_.end()) << "Cannot find layer: " << layerName;
-  return it->second->getOutputValue();
+Argument NeuralNetwork::getLayerOutput(const std::string& layerName) {
+  return getLayer(layerName)->getOutput();
 }
+
 void NeuralNetwork::onPassEnd() {
   for (auto& layer : layers_) {
     layer->onPassEnd();

paddle/gserver/gradientmachines/NeuralNetwork.h

@@ -87,7 +87,8 @@ public:
 
   virtual void backward(const UpdateCallback& callback = nullptr);
 
-  MatrixPtr getLayerOutput(const std::string& layerName);
+  virtual Argument getLayerOutput(const std::string& layerName);
+
   const LayerPtr& getLayer(const std::string& layerName) const {
     auto it = layerMap_.find(layerName);
     CHECK(it != layerMap_.end()) << "Unknown layer " << layerName;

paddle/gserver/layers/CosSimLayer.cpp

@@ -42,7 +42,7 @@ void CosSimLayer::forward(PassType passType) {
   /* malloc memory for the output_ if necessary */
   int batchSize = getInputValue(0)->getHeight();
   int size = getSize();
-  CHECK_EQ(forward_.size(), 1) << "Only one forward function needed";
+  CHECK_EQ(forward_.size(), 1UL) << "Only one forward function needed";
 
   {
     REGISTER_TIMER_INFO("CosFwResetTimer", getName().c_str());
@@ -68,7 +68,7 @@ void CosSimLayer::forward(PassType passType) {
 void CosSimLayer::backward(const UpdateCallback& callback) {
   /* activation */ {
     REGISTER_TIMER_INFO("CosBpAtvTimer", getName().c_str());
-    CHECK_EQ(backward_.size(), 1) << "Only one backward function needed";
+    CHECK_EQ(backward_.size(), 1UL) << "Only one backward function needed";
 
     const auto outG = this->getOutputGrad();
     const auto outV = this->getOutputValue();

paddle/gserver/layers/CosSimVecMatLayer.cpp

@@ -112,7 +112,7 @@ bool CosSimVecMatLayer::init(const LayerMap& layerMap,
 
 void CosSimVecMatLayer::forward(PassType passType) {
   Layer::forward(passType);
-  CHECK_EQ(forward_.size(), 1) << "Only one forward function needed";
+  CHECK_EQ(forward_.size(), 1UL) << "Only one forward function needed";
 
   MatrixPtr inV0 = getInputValue(0);
   MatrixPtr inV1 = getInputValue(1);
@@ -145,7 +145,7 @@ void CosSimVecMatLayer::forward(PassType passType) {
 }
 
 void CosSimVecMatLayer::backward(const UpdateCallback& callback) {
-  CHECK_EQ(backward_.size(), 1) << "Only one forward function needed";
+  CHECK_EQ(backward_.size(), 1UL) << "Only one forward function needed";
 
   MatrixPtr inV0 = getInputValue(0);
   MatrixPtr inV1 = getInputValue(1);

paddle/math/tests/test_RowBuffer.cpp

@@ -17,10 +17,10 @@ limitations under the License. */
 
 TEST(RowBuffer, testAutoGrow) {
   paddle::RowBuffer buf(128);
-  ASSERT_EQ(128, buf.getWidth());
+  ASSERT_EQ(128UL, buf.getWidth());
   ASSERT_TRUE(buf.isAutoGrowth());
   buf.resize(2);
-  ASSERT_EQ(2, buf.getRowCount());
+  ASSERT_EQ(2UL, buf.getRowCount());
   for (size_t i = 0; i < buf.getWidth() * 2; ++i) {
     buf.data()[i] = i;
   }
@@ -35,7 +35,7 @@ TEST(RowBuffer, testAutoGrow) {
     data[i] = i;
   }
 
-  ASSERT_EQ(3, buf.getRowCount());
+  ASSERT_EQ(3UL, buf.getRowCount());
   for (size_t i = 0; i < buf.getRowCount() - 1; ++i) {
     for (size_t j = 0; j < buf.getWidth(); ++j) {
       ASSERT_NEAR(i * buf.getWidth() + j, buf.get(i)[j], 1e-5);
@@ -51,7 +51,7 @@ TEST(RowBuffer, testWithMemBuf) {
       std::make_shared<paddle::CpuMemoryHandle>(128 * 2 * sizeof(real));
   paddle::RowBuffer buf(mem, 128);
   ASSERT_TRUE(!buf.isAutoGrowth());
-  ASSERT_EQ(2, buf.getRowCount());
+  ASSERT_EQ(2UL, buf.getRowCount());
   for (size_t i = 0; i < buf.getWidth() * 2; ++i) {
     buf.data()[i] = i;
   }

paddle/py_paddle/dataprovider_converter.py

@@ -23,7 +23,8 @@ __all__ = ['DataProviderConverter']
 class IScanner(object):
     def __init__(self, input_type, pos):
         self.input_type = input_type
-        assert isinstance(self.input_type, dp2.InputType)
+        if not isinstance(self.input_type, dp2.InputType):
+            raise ValueError("input type should be dataprovider2.InputType")
         self.pos = pos
 
     def scan(self, dat):
@@ -50,7 +51,6 @@ class DenseScanner(IScanner):
 
     def finish_scan(self, argument):
         assert isinstance(argument, swig_paddle.Arguments)
-        assert isinstance(self.input_type, dp2.InputType)
         if self.__mat__.dtype != numpy.float32:
             self.__mat__ = self.__mat__.astype(numpy.float32)
         m = swig_paddle.Matrix.createDenseFromNumpy(self.__mat__, True, False)
@@ -63,7 +63,6 @@ class SparseBinaryScanner(IScanner):
         self.__rows__ = [0]
         self.__cols__ = []
         self.__height__ = 0
-        self.__nnz__ = 0
         self.__value__ = []
 
     def scan(self, dat):
@@ -76,7 +75,6 @@ class SparseBinaryScanner(IScanner):
 
     def finish_scan(self, argument):
         assert isinstance(argument, swig_paddle.Arguments)
-        assert isinstance(self.input_type, dp2.InputType)
         m = swig_paddle.Matrix.createSparse(self.__height__,
                                             self.input_type.dim,
                                             len(self.__cols__),

paddle/py_paddle/util.py

@@ -208,7 +208,7 @@ def __monkeypatch_gradient_machine__():
 
         output = dict()
         for name in layerNames:
-            output[name] = __matrix_to_numpy__(self.getLayerOutput(name))
+            output[name] = __arguments_to_numpy__(0, self.getLayerOutput(name))
         return output
 
     swig_paddle.GradientMachine.getLayerOutputs = getLayerOutputs

paddle/scripts/docker/Dockerfile

@@ -10,28 +10,30 @@ RUN apt-get update && \
     apt-get install -y wget unzip tar xz-utils bzip2 gzip coreutils && \
     apt-get install -y curl sed grep graphviz libjpeg-dev zlib1g-dev && \
     apt-get install -y python-numpy python-matplotlib gcc g++ gfortran && \
-    apt-get install -y automake clang-3.8 llvm-3.8 libclang-3.8-dev && \
+    apt-get install -y automake && \
     apt-get clean -y
 
 RUN pip install --upgrade pip && \
-    pip install -U protobuf && \
+    pip install -U "protobuf==3.1.0" && \
     pip install -U wheel pillow BeautifulSoup && \
     pip install -U docopt PyYAML sphinx && \
     pip install -U sphinx_rtd_theme recommonmark jupyter
 
 RUN curl -sSL https://cmake.org/files/v3.4/cmake-3.4.1.tar.gz | tar -xz && \
-    cd cmake-3.4.1 && ./bootstrap && make -j4 && make install && \
+    cd cmake-3.4.1 && ./bootstrap && make -j `nproc` && make install && \
     cd .. && rm -rf cmake-3.4.1
 
+ARG BUILD_WOBOQ
 ARG BUILD_AND_INSTALL
 ARG WITH_AVX
 ARG WITH_DOC
 ARG WITH_STYLE_CHECK
 
+ENV BUILD_WOBOQ=${BUILD_WOBOQ:-OFF}
 ENV BUILD_AND_INSTALL=${BUILD_AND_INSTALL:-OFF}
 ENV WITH_GPU=OFF
 ENV WITH_AVX=${WITH_AVX:-ON}
-ENV WITH_DOC=${WITH_DOC:-ON}
+ENV WITH_DOC=${WITH_DOC:-OFF}
 ENV WITH_STYLE_CHECK=${WITH_STYLE_CHECK:-OFF}
 
 RUN mkdir /paddle

paddle/scripts/docker/Dockerfile.gpu

@@ -10,28 +10,30 @@ RUN apt-get update && \
     apt-get install -y wget unzip tar xz-utils bzip2 gzip coreutils && \
     apt-get install -y curl sed grep graphviz libjpeg-dev zlib1g-dev && \
     apt-get install -y python-numpy python-matplotlib gcc g++ gfortran && \
-    apt-get install -y automake clang-3.8 llvm-3.8 libclang-3.8-dev && \
+    apt-get install -y automake && \
     apt-get clean -y
 
 RUN pip install --upgrade pip && \
-    pip install -U protobuf && \
+    pip install -U "protobuf==3.1.0" && \
     pip install -U wheel pillow BeautifulSoup && \
     pip install -U docopt PyYAML sphinx && \
     pip install -U sphinx_rtd_theme recommonmark jupyter
 
 RUN curl -sSL https://cmake.org/files/v3.4/cmake-3.4.1.tar.gz | tar -xz && \
-    cd cmake-3.4.1 && ./bootstrap && make -j4 && make install && \
+    cd cmake-3.4.1 && ./bootstrap && make -j `nproc` && make install && \
     cd .. && rm -rf cmake-3.4.1
 
+ARG BUILD_WOBOQ
 ARG BUILD_AND_INSTALL
 ARG WITH_AVX
 ARG WITH_DOC
 ARG WITH_STYLE_CHECK
 
+ENV BUILD_WOBOQ=${BUILD_WOBOQ:-OFF}
 ENV BUILD_AND_INSTALL=${BUILD_AND_INSTALL:-OFF}
 ENV WITH_GPU=ON
 ENV WITH_AVX=${WITH_AVX:-ON}
-ENV WITH_DOC=${WITH_DOC:-ON}
+ENV WITH_DOC=${WITH_DOC:-OFF}
 ENV WITH_STYLE_CHECK=${WITH_STYLE_CHECK:-OFF}
 
 RUN mkdir /paddle

paddle/scripts/docker/build.sh

@@ -11,7 +11,7 @@ set -e
 # If Dockerfile.* sets BUILD_AND_INSTALL to 'ON', it would have copied
 # source tree to /paddle, and this scripts should build it into
 # /paddle/build.
-if [[ ${BUILD_AND_INSTALL:-ON} == 'ON' ]]; then
+if [[ ${BUILD_AND_INSTALL:-OFF} == 'ON' ]]; then
     if [[ ${WITH_GPU:-OFF} == 'ON' ]]; then
 	ln -s /usr/lib/x86_64-linux-gnu/libcudnn.so /usr/lib/libcudnn.so
     fi
@@ -19,7 +19,7 @@ if [[ ${BUILD_AND_INSTALL:-ON} == 'ON' ]]; then
     mkdir -p /paddle/build # -p means no error if exists
     cd /paddle/build
     cmake .. \
-	  -DWITH_DOC=ON \
+	  -DWITH_DOC=${WITH_DOC:-OFF} \
 	  -DWITH_GPU=${WITH_GPU:-OFF} \
 	  -DWITH_AVX=${WITH_AVX:-OFF} \
 	  -DWITH_SWIG_PY=ON \
@@ -29,6 +29,8 @@ if [[ ${BUILD_AND_INSTALL:-ON} == 'ON' ]]; then
     make -j `nproc`
     make install
 
+    if [[ ${BUILD_WOBOQ:-OFF} == 'ON' ]]; then
+        apt-get install -y clang-3.8 llvm-3.8 libclang-3.8-dev 
         # Install woboq_codebrowser.
         git clone https://github.com/woboq/woboq_codebrowser /woboq
         cd /woboq
@@ -49,8 +51,10 @@ if [[ ${BUILD_AND_INSTALL:-ON} == 'ON' ]]; then
         /woboq/indexgenerator/codebrowser_indexgenerator $WOBOQ_OUT
         cd /woboq
         make clean
+    fi
 
-    pip install /usr/local/opt/paddle/share/wheels/*.whl
+    pip install /usr/local/opt/paddle/share/wheels/py_paddle*linux*.whl
+    pip install /usr/local/opt/paddle/share/wheels/paddle*.whl
     paddle version
 fi
 

python/CMakeLists.txt

@@ -25,6 +25,7 @@ add_custom_target(paddle_python ALL DEPENDS
 
 add_subdirectory(paddle/trainer_config_helpers/tests)
 add_subdirectory(paddle/reader/tests)
+add_subdirectory(paddle/v2/tests)
 
 install(DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/dist/
     DESTINATION opt/paddle/share/wheels

python/paddle/reader/__init__.py

@@ -21,3 +21,5 @@
 #
 #     r = paddle.reader.buffered(paddle.reader.creator.text("hello.txt"))
 from decorator import *
+
+import creator