Merge branch 'develop' of github.com:baidu/Paddle into feature/enhance_evaluator

Author: reyoung

paddle/framework/backward.cc

@@ -522,7 +522,7 @@ ParamGradInfoMap AppendBackward(
       new OpDescBind("fill_constant", {}, {{"Out", {fill_one_op_out}}},
                      {{"shape", std::vector<int>{1}},
                       {"value", static_cast<float>(1.0)},
-                      {"data_type", target.GetDataType()}}));
+                      {"dtype", target.GetDataType()}}));
   // infer var type of fill_one_op
   fill_one_op->InferVarType(root_block);
 

paddle/framework/tensor_array.cc

@@ -302,7 +302,7 @@ LoDTensor TensorArray::Stack() const {
 
   const auto& first_dims = values_.front().dims();
   // check all the values have the same shape
-  // TODO(superjom) check the same dtypes
+  // TODO(superjom) check the same data_type
   for (size_t idx = 1; idx < size(); idx++) {
     const auto& value_dims = values_[idx].dims();
     PADDLE_ENFORCE_EQ(first_dims, value_dims);

paddle/memory/CMakeLists.txt

@@ -1,6 +1,6 @@
 add_subdirectory(detail)
 
-cc_library(memory SRCS memory.cc DEPS place)
+cc_library(memory SRCS memory.cc DEPS place enforce)
 cc_library(memcpy SRCS memcpy.cc)
 
 cc_library(paddle_memory

paddle/operators/bilinear_tensor_product_op.cc

@@ -77,11 +77,19 @@ class BilinearTensorProductOpMaker : public framework::OpProtoAndCheckerMaker {
     AddOutput("Out", "The output of bilinear_tensor_product operator.");
     AddComment(R"DOC(
 Bilinear Tensor Product operator.
-Given input X and Y, a 3D tensor weight, and bias. Each column of the
-output is computed by one slice i = 1, . . . , k of the tensor:
-
-    M =  (X W_i) \cdot Y
-    Out_i = \sum_i {M_i} + Bias_i
+Given input X and Y, a 3D tensor Weight and a Bias. Each column of the
+Output is computed by one slice $i = 1, . . . , k$ of the tensor:
+
+$$
+M =  (X W_i) * Y \\
+Out_i = \sum_j {M_j} + Bias_i
+$$
+
+Where $W_i$ is the $i$-th slice of Input(Weight);
+      $M_j$ is the $j$-th column of $M$;
+      $Out_i$ is the $i$-th column of Output(Out);
+      $Bias_i$ is a column vector, each element of it is equal to
+        the $i$-th element of $Bias$;
 
 )DOC");
   }

paddle/operators/cast_op.cc

@@ -25,8 +25,8 @@ class CastOpProtoMaker : public framework::OpProtoAndCheckerMaker {
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("X", "The input tensor of cast op");
     AddOutput("Out", "The output tensor of cast op");
-    AddAttr<int>("out_data_type", "output data type");
-    AddAttr<int>("in_data_type", "input data type");
+    AddAttr<int>("out_dtype", "output data type");
+    AddAttr<int>("in_dtype", "input data type");
     AddComment(R"DOC(
 Cast Operator.
 
@@ -58,8 +58,8 @@ class CastOpGradMaker : public framework::SingleGradOpDescMaker {
     grad->SetType("cast");
     grad->SetInput("X", OutputGrad("Out"));
     grad->SetOutput("Out", InputGrad("X"));
-    grad->SetAttr("out_data_type", GetAttr("in_data_type"));
-    grad->SetAttr("in_data_type", GetAttr("out_data_type"));
+    grad->SetAttr("out_dtype", GetAttr("in_dtype"));
+    grad->SetAttr("in_dtype", GetAttr("out_dtype"));
     return std::unique_ptr<framework::OpDescBind>(grad);
   }
 };

paddle/operators/cast_op.h

@@ -55,7 +55,7 @@ class CastOpKernel : public framework::OpKernel<InT> {
     auto* in = context.Input<framework::Tensor>("X");
     auto* out = context.Output<framework::Tensor>("Out");
     framework::VisitDataType(
-        static_cast<framework::DataType>(context.Attr<int>("out_data_type")),
+        static_cast<framework::DataType>(context.Attr<int>("out_dtype")),
         CastOpFunctor<Place, InT>(in, out, context.device_context()));
   }
 };

paddle/operators/fill_constant_batch_size_like_op.cc

@@ -52,7 +52,7 @@ class FillConstantBatchSizeLikeOp : public framework::OperatorWithKernel {
   framework::OpKernelType GetKernelType(
       const framework::ExecutionContext &ctx) const override {
     return framework::OpKernelType(
-        static_cast<framework::DataType>(ctx.Attr<int>("data_type")),
+        static_cast<framework::DataType>(ctx.Attr<int>("dtype")),
         ctx.device_context());
   }
 };
@@ -63,7 +63,7 @@ class FillConstantBatchSizeLikeOpMaker
   FillConstantBatchSizeLikeOpMaker(framework::OpProto *proto,
                                    framework::OpAttrChecker *op_checker)
       : framework::OpProtoAndCheckerMaker(proto, op_checker) {
-    AddAttr<int>("data_type",
+    AddAttr<int>("dtype",
                  "(int, default 5 (FP32)) "
                  "Output data type")
         .SetDefault(framework::DataType::FP32);

paddle/operators/fill_constant_op.cc

@@ -34,7 +34,7 @@ class FillConstantOp : public framework::OperatorBase {
   using framework::OperatorBase::OperatorBase;
   void Run(const framework::Scope &scope,
            const platform::DeviceContext &dev_ctx) const override {
-    auto data_type = static_cast<framework::DataType>(Attr<int>("data_type"));
+    auto data_type = static_cast<framework::DataType>(Attr<int>("dtype"));
     auto value = Attr<float>("value");
     auto force_cpu = Attr<bool>("force_cpu");
     auto &out =
@@ -55,7 +55,7 @@ class FillConstantOpMaker : public framework::OpProtoAndCheckerMaker {
   FillConstantOpMaker(framework::OpProto *proto,
                       framework::OpAttrChecker *op_checker)
       : framework::OpProtoAndCheckerMaker(proto, op_checker) {
-    AddAttr<int>("data_type",
+    AddAttr<int>("dtype",
                  "(int, default 5 (FP32)) "
                  "Output data type")
         .SetDefault(framework::DataType::FP32);

paddle/operators/gaussian_random_op.cc

@@ -60,7 +60,7 @@ class GaussianRandomOp : public framework::OperatorWithKernel {
   framework::OpKernelType GetKernelType(
       const framework::ExecutionContext& ctx) const override {
     return framework::OpKernelType(
-        static_cast<framework::DataType>(ctx.Attr<int>("data_type")),
+        static_cast<framework::DataType>(ctx.Attr<int>("dtype")),
         ctx.device_context());
   }
 };
@@ -88,7 +88,7 @@ class GaussianRandomOpMaker : public framework::OpProtoAndCheckerMaker {
                  "Random seed of generator."
                  "0 means use system wide seed.")
         .SetDefault(0);
-    AddAttr<int>("data_type",
+    AddAttr<int>("dtype",
                  "(int, default 5(FP32)) "
                  "Output data type.")
         .SetDefault(framework::DataType::FP32);

paddle/operators/huber_loss_op.cc

@@ -70,11 +70,18 @@ input value and Y as the target value. Huber loss can evaluate the fitness of
 X to Y. Different from MSE loss, Huber loss is more robust for outliers. The
 shape of X and Y are [batch_size, 1]. The equation is:
 
-L_{\delta}(y, f(x)) =
+$$
+Out_{\delta}(X, Y)_i =
 \begin{cases}
-0.5 * (y - f(x))^2, \quad |y - f(x)| \leq \delta \\
-\delta * (|y - f(x)| - 0.5 * \delta),   \quad otherwise
+0.5 * (Y_i - X_i)^2,
+\quad |Y_i - X_i| \leq \delta \\
+\delta * (|Y_i - X_i| - 0.5 * \delta),
+\quad otherwise
 \end{cases}
+$$
+
+In the above equation, $Out_\delta(X, Y)_i$, $X_i$ and $Y_i$ represent the ith
+element of Out, X and Y.
 
 )DOC");
   }