diff --git a/AUTHORS.md b/AUTHORS.md index 11f227be7148d8d6e055538347a8c31679406c84..8c4a113fc276783c945867ceae9612339b7f0bbc 100644 --- a/AUTHORS.md +++ b/AUTHORS.md @@ -22,6 +22,7 @@ | jczaja | Jacek Czaja | | JiayiFeng | Jia-Yi Feng | | kbinias | Krzysztof Binias | +| kexinzhao | Ke-Xin Zhao | | kuke | Yi-Bing Liu | | lcy-seso | Ying Cao | | lipeng-unisound | Peng Li | diff --git a/cmake/inference_lib.cmake b/cmake/inference_lib.cmake index 236a55d332a91c88d1c5515e7aca4142930a079f..cd44fe2542bfa8c53721d61b70778226e640d375 100644 --- a/cmake/inference_lib.cmake +++ b/cmake/inference_lib.cmake @@ -39,7 +39,7 @@ function(copy TARGET) message(FATAL_ERROR "${TARGET} source numbers are not equal to destination numbers") endif() math(EXPR len "${copy_lib_SRCS_len} - 1") - + add_custom_target(${TARGET} DEPENDS ${copy_lib_DEPS}) foreach(index RANGE ${len}) list(GET copy_lib_SRCS ${index} src) @@ -155,6 +155,15 @@ copy(inference_lib DEPS paddle_fluid_shared paddle_fluid DSTS ${dst_dir}/${module} ${dst_dir}/${module} ) +if(WITH_CONTRIB) + set(contrib_dst_dir "${FLUID_INSTALL_DIR}/contrib/inference") + copy(contrib_inference_lib DEPS paddle_inference_api + SRCS ${PADDLE_SOURCE_DIR}/paddle/contrib/inference/paddle_inference_api.h + ${PADDLE_BINARY_DIR}/paddle/contrib/inference/libpaddle_inference_api.* + DSTS ${contrib_dst_dir} ${contrib_dst_dir} + ) +endif() + set(module "platform") copy(platform_lib DEPS profiler_py_proto SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/dynload/*.h ${src_dir}/${module}/details/*.h diff --git a/doc/fluid/api/layers.rst b/doc/fluid/api/layers.rst index e5ced9c04c3f702733635ad0397c8c52ec4b3970..8d1c9247b1250703ee605edd21b1cd8fe74a9787 100644 --- a/doc/fluid/api/layers.rst +++ b/doc/fluid/api/layers.rst @@ -342,6 +342,12 @@ conv2d .. autofunction:: paddle.fluid.layers.conv2d :noindex: +conv3d +------ + +.. autofunction:: paddle.fluid.layers.conv3d + :noindex: + sequence_pool ------------- @@ -366,6 +372,12 @@ pool2d .. autofunction:: paddle.fluid.layers.pool2d :noindex: +pool3d +------ + +.. autofunction:: paddle.fluid.layers.pool3d + :noindex: + batch_norm ---------- @@ -384,6 +396,13 @@ conv2d_transpose .. autofunction:: paddle.fluid.layers.conv2d_transpose :noindex: +conv3d_transpose +---------------- + +.. autofunction:: paddle.fluid.layers.conv2d_transpose + :noindex: + + sequence_expand --------------- diff --git a/doc/v2/dev/contribute_to_paddle_cn.md b/doc/v2/dev/contribute_to_paddle_cn.md index add06e42f1bbd221b48eb83e4e84d4a7c89e7483..3244eedf918b93f9351258f1218dfb2d507c1a9c 100644 --- a/doc/v2/dev/contribute_to_paddle_cn.md +++ b/doc/v2/dev/contribute_to_paddle_cn.md @@ -104,7 +104,7 @@ no changes added to commit (use "git add" and/or "git commit -a") ➜ docker run -it -v $(pwd):/paddle paddle:latest-dev bash -c "cd /paddle/build && ctest" ``` -关于构建和测试的更多信息，请参见[这篇文档](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/getstarted/build_and_install/docker_install_cn.rst)。 +关于构建和测试的更多信息，请参见[使用Docker安装运行](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/v2/build_and_install/docker_install_cn.rst)。 ## 提交（commit） diff --git a/paddle/contrib/inference/CMakeLists.txt b/paddle/contrib/inference/CMakeLists.txt index 277b0b175b29f682eed5a6584867ffa239d9d081..0f56d648b1939e1d6af3368bb2423477a3b638fc 100644 --- a/paddle/contrib/inference/CMakeLists.txt +++ b/paddle/contrib/inference/CMakeLists.txt @@ -50,7 +50,7 @@ cc_test(test_paddle_inference_api inference_api_test(test_paddle_inference_api_impl ARGS test_word2vec test_image_classification) -if (WITH_ANAKIN) +if (WITH_ANAKIN AND WITH_TESTING) # only needed in CI # Due to Anakin do not have official library releases and the versions of protobuf and cuda do not match Paddle's, # so anakin library will not be merged to our official inference library. To use anakin prediction API, one need to # compile the libinference_anakin_api.a and compile with anakin.so. diff --git a/paddle/contrib/tape/CMakeLists.txt b/paddle/contrib/tape/CMakeLists.txt index 470abf25c6c63335df67ce152dc99be92d131cc6..5450359d859de93ca19c56422f1243c7f445aff7 100644 --- a/paddle/contrib/tape/CMakeLists.txt +++ b/paddle/contrib/tape/CMakeLists.txt @@ -17,7 +17,7 @@ if(APPLE) set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-error=pessimizing-move") endif(APPLE) -cc_library(tape_variable SRCS variable.cc DEPS ${FLUID_CORE_MODULES} device_context) +cc_library(tape_variable SRCS variable.cc DEPS ${FLUID_CORE_MODULES} device_context framework_proto proto_desc operator) cc_library(tape SRCS tape.cc DEPS ${FLUID_CORE_MODULES} ${GLOB_OP_LIB} tape_variable) cc_test(test_tape diff --git a/paddle/fluid/operators/activation_op.cc b/paddle/fluid/operators/activation_op.cc index af1d85047e519df6766b2139a0445ae9dc5945e2..c73482eb12e882fe15a595ad485ae688db346803 100644 --- a/paddle/fluid/operators/activation_op.cc +++ b/paddle/fluid/operators/activation_op.cc @@ -271,18 +271,18 @@ class HardShrinkOpMaker : public framework::OpProtoAndCheckerMaker { void Make() override { AddInput("X", "Input of HardShrink operator"); AddOutput("Out", "Output of HardShrink operator"); - AddAttr("threshold", "The value of threshold for HardShrink") + AddAttr("threshold", + "The value of threshold for HardShrink. [default: 0.5]") .SetDefault(0.5f); AddComment(R"DOC( -HardShrink Activation Operator. +:strong:`HardShrink activation operator` -$$ -out = \begin{cases} - x, \text{if } x > \lambda \\ - x, \text{if } x < -\lambda \\ - 0, \text{otherwise} - \end{cases} -$$ +.. math:: + out = \begin{cases} + x, \text{if } x > \lambda \\ + x, \text{if } x < -\lambda \\ + 0, \text{otherwise} + \end{cases} )DOC"); } @@ -394,18 +394,18 @@ class ThresholdedReluOpMaker : public framework::OpProtoAndCheckerMaker { void Make() override { AddInput("X", "Input of ThresholdedRelu operator"); AddOutput("Out", "Output of ThresholdedRelu operator"); - AddAttr("threshold", "The threshold location of activation") + AddAttr("threshold", + "The threshold location of activation. [default 1.0].") .SetDefault(1.0f); AddComment(R"DOC( -ThresholdedRelu Activation Operator. +:strong:`ThresholdedRelu activation operator` -$$ -out = \begin{cases} - x, \text{if } x > threshold \\ - 0, \text{otherwise} - \end{cases} -$$ +.. math:: + out = \begin{cases} + x, \text{if } x > threshold \\ + 0, \text{otherwise} + \end{cases} )DOC"); } }; diff --git a/paddle/fluid/operators/compare_op.cc b/paddle/fluid/operators/compare_op.cc index 3a4819f3dec9704a4a7c8910dd22e80fda082335..f40b1ba338d429c248103eeb930ac7e1bb690218 100644 --- a/paddle/fluid/operators/compare_op.cc +++ b/paddle/fluid/operators/compare_op.cc @@ -23,30 +23,26 @@ class CompareOpProtoMaker : public framework::OpProtoAndCheckerMaker { public: void Make() override { OpComment comment; - AddInput("X", - string::Sprintf("(LoDTensor) the left hand operand of %s operator", - comment.type)); - AddInput("Y", string::Sprintf( - "(LoDTensor) the right hand operand of %s operator", - comment.type)); + AddInput("X", string::Sprintf("the left hand operand of %s operator", + comment.type)); + AddInput("Y", string::Sprintf("the right hand operand of %s operator", + comment.type)); AddAttr("force_cpu", - "(bool, default false) Force fill output variable to cpu " + "Force fill output variable to cpu " "memory. Otherwise, fill output variable to the running " - "device") - .SetDefault(false); - AddOutput("Out", string::Sprintf( - "(LoDTensor) n-dim bool tensor. Each element is %s", - comment.equation)); - AddComment(string::Sprintf(R"DOC(%s Operator - + "device [default true].") + .SetDefault(true); + AddOutput("Out", string::Sprintf("n-dim bool tensor. Each element is %s", + comment.equation)); + AddComment(string::Sprintf(R"DOC( It operates element-wise on X and Y, and returns the Out. Each of them is a N-dim tensor. X and Y could be any type. The each element of the Out tensor is -calculated by %s +calculated by $%s$ )DOC", - comment.type, comment.equation)); - AddAttr("axis", - "(int, default -1). The start dimension index " - "for broadcasting Y onto X.") + comment.equation)); + AddAttr( + "axis", + "The start dimension index for broadcasting Y onto X. [default -1]") .SetDefault(-1) .EqualGreaterThan(-1); } diff --git a/paddle/fluid/operators/concat_op.cc b/paddle/fluid/operators/concat_op.cc index 38337f9aa52435c445420047957500d21069506a..c72405593788493e10a1293b0c722e2d11c6e312 100644 --- a/paddle/fluid/operators/concat_op.cc +++ b/paddle/fluid/operators/concat_op.cc @@ -107,7 +107,13 @@ REGISTER_OPERATOR(concat, ops::ConcatOp, ops::ConcatOpMaker, false> /* set false to disable empty grad */); REGISTER_OPERATOR(concat_grad, ops::ConcatOpGrad); REGISTER_OP_CPU_KERNEL( - concat, ops::ConcatKernel); + concat, ops::ConcatKernel, + ops::ConcatKernel, + ops::ConcatKernel, + ops::ConcatKernel); REGISTER_OP_CPU_KERNEL( concat_grad, - ops::ConcatGradKernel); + ops::ConcatGradKernel, + ops::ConcatGradKernel, + ops::ConcatGradKernel, + ops::ConcatGradKernel); diff --git a/paddle/fluid/operators/concat_op.cu.cc b/paddle/fluid/operators/concat_op.cu.cc index 590eca9d066ff7549939e62ddbfedc8ab76bb5e7..8e38e5231fbf6955ff8a9680a241a4a4ba1b924d 100644 --- a/paddle/fluid/operators/concat_op.cu.cc +++ b/paddle/fluid/operators/concat_op.cu.cc @@ -15,7 +15,13 @@ limitations under the License. */ #include "paddle/fluid/operators/concat_op.h" namespace ops = paddle::operators; REGISTER_OP_CUDA_KERNEL( - concat, ops::ConcatKernel); + concat, ops::ConcatKernel, + ops::ConcatKernel, + ops::ConcatKernel, + ops::ConcatKernel); REGISTER_OP_CUDA_KERNEL( concat_grad, - ops::ConcatGradKernel); + ops::ConcatGradKernel, + ops::ConcatGradKernel, + ops::ConcatGradKernel, + ops::ConcatGradKernel); diff --git a/paddle/fluid/operators/cumsum_op.cc b/paddle/fluid/operators/cumsum_op.cc index 92bb835e8f18e17ae1355fdec29f43b8ffb70460..5302b822d6b9f232e9ccd0d03cc549d7d5044ebf 100644 --- a/paddle/fluid/operators/cumsum_op.cc +++ b/paddle/fluid/operators/cumsum_op.cc @@ -30,19 +30,19 @@ class CumOp : public framework::OperatorWithKernel { class CumsumOpMaker : public framework::OpProtoAndCheckerMaker { public: void Make() override { - AddInput("X", "Input of Cumsum operator"); - AddOutput("Out", "Output of Cumsum operator"); + AddInput("X", "Input of cumsum operator"); + AddOutput("Out", "Output of cumsum operator"); AddAttr("axis", - "(int, default -1). The dimenstion to accumulate along. " - "-1 means the last dimenstion") + "The dimenstion to accumulate along. -1 means the last " + "dimenstion [default -1].") .SetDefault(-1) .EqualGreaterThan(-1); AddAttr("exclusive", - "bool, default false). Whether to perform exclusive cumsum") + "Whether to perform exclusive cumsum. [default false].") .SetDefault(false); AddAttr("reverse", - "bool, default false). If true, the cumsum is performed in " - "the reversed direction") + "If true, the cumsum is performed in the reversed direction. " + "[default false].") .SetDefault(false); AddComment(R"DOC( The cumulative sum of the elements along a given axis. diff --git a/paddle/fluid/operators/get_places_op.cc b/paddle/fluid/operators/get_places_op.cc index eafc364a15fa17cc5107bba737b0b44e712b0bef..db6ff7825690176ded0ab957764ed8411d3cd804 100644 --- a/paddle/fluid/operators/get_places_op.cc +++ b/paddle/fluid/operators/get_places_op.cc @@ -85,7 +85,7 @@ class GetPlacesOpProtoMaker : public framework::OpProtoAndCheckerMaker { .InEnum({"CUDA", "CPU", "AUTO"}) .SetDefault("AUTO"); AddComment(R"DOC( -Returns a list of places based on flags. The list will be used for parallel +Returns a list of places based on arguments. The list will be used for parallel execution. )DOC"); } diff --git a/paddle/fluid/operators/layer_norm_op.cc b/paddle/fluid/operators/layer_norm_op.cc index ab097d31e9ab5eafa788539170e7e405df697625..14ce1da2e97186a50ed8bd52223a500c4c57b328 100644 --- a/paddle/fluid/operators/layer_norm_op.cc +++ b/paddle/fluid/operators/layer_norm_op.cc @@ -62,36 +62,33 @@ class LayerNormOp : public framework::OperatorWithKernel { class LayerNormOpMaker : public framework::OpProtoAndCheckerMaker { public: void Make() override { - AddInput("X", "(LoDTensor) The input tensor."); + AddInput("X", "The input tensor."); AddInput("Scale", - "(Tensor, optional) Scale is a 1-dimensional tensor of size " + "(optional) Scale is a 1-dimensional tensor of size " "H(`begin_norm_axis` splits the tensor(`X`) to a matrix [N,H])." "It is applied to the output.") .AsDispensable(); AddInput("Bias", - "(Tensor, optional) Bias is a 1-dimensional tensor of size " + "(optional) Bias is a 1-dimensional tensor of size " "H(`begin_norm_axis` splits the tensor(`X`) to a matrix [N,H])." "It is applied to the output.") .AsDispensable(); - AddOutput("Y", "(LoDTensor) Result after normalization."); - AddOutput("Mean", "(Tensor) Mean of the current mini batch.") - .AsIntermediate(); - AddOutput("Variance", "(Tensor) Variance of the current mini batch.") + AddOutput("Y", "Result after normalization."); + AddOutput("Mean", "Mean of the current mini batch.").AsIntermediate(); + AddOutput("Variance", "Variance of the current mini batch.") .AsIntermediate(); AddAttr("epsilon", - "(float, default 1e-5) Constant for " - "numerical stability") + "Constant for numerical stability [default 1e-5].") .SetDefault(1e-5) .AddCustomChecker([](const float &epsilon) { PADDLE_ENFORCE(epsilon >= 0.0f && epsilon <= 0.001f, "'epsilon' should be between 0.0 and 0.001."); }); AddAttr("begin_norm_axis", - "(int default:1), the " - "axis of `begin_norm_axis ... Rank(X) - 1` will be " + "the axis of `begin_norm_axis ... Rank(X) - 1` will be " "normalized. `begin_norm_axis` splits the tensor(`X`) to a " - "matrix [N,H].") + "matrix [N,H]. [default 1].") .SetDefault(1) .AddCustomChecker([](const int &begin_norm_axis) { PADDLE_ENFORCE_GT(begin_norm_axis, 0, @@ -99,10 +96,14 @@ class LayerNormOpMaker : public framework::OpProtoAndCheckerMaker { }); AddComment(R"DOC( -Layer Normalization. -Layer Norm has been implemented as discussed in the paper: -https://arxiv.org/abs/1607.06450 -... +Assume feature vectors exist on dimensions +:attr:`begin_norm_axis ... rank(input)` and calculate the moment statistics +along these dimensions for each feature vector :math:`a` with size +:math:`H`, then normalize each feature vector using the corresponding +statistics. After that, apply learnable gain and bias on the normalized +tensor to scale and shift if :attr:`scale` and :attr:`shift` are set. + +Refer to `Layer Normalization `_ )DOC"); } }; diff --git a/paddle/fluid/operators/multiplex_op.cc b/paddle/fluid/operators/multiplex_op.cc index a4363fd25d57edb5c2509904a1f55634832613be..18ad46cb5eeeab2169136e40cebdaa53c0bfd587 100644 --- a/paddle/fluid/operators/multiplex_op.cc +++ b/paddle/fluid/operators/multiplex_op.cc @@ -62,26 +62,46 @@ class MultiplexOp : public framework::OperatorWithKernel { class MultiplexOpMaker : public framework::OpProtoAndCheckerMaker { public: void Make() override { - AddInput("Ids", "The index tensor of multiplex operator."); - AddInput("X", "The candidate tensors of multiplex operator.") + AddInput("Ids", + "Tensor, index variable which is a 2-D tensor with shape " + "[M, 1] where M is the batch size."); + AddInput("X", + "A list of variables to gather from. All variables have the same " + "shape and the rank is at least 2.") .AsDuplicable(); AddOutput("Out", "The output tensor of multiplex operator."); AddComment(R"DOC( -Multiplex Operator. - -Multiplex multiple tensors according to the index provided by the index tensor. - -Ids: the index tensor. -X[0 : N - 1]: the candidate tensors for output (N >= 2). -For each index i from 0 to batchSize - 1, the output is the i-th row of the +Referring to the given index variable, this layer selects rows from the +input variables to construct a multiplex variable. Assuming that there are +:math:`m` input variables and :math:`I_i` represents the i-th input +variable and :math:`i` is in [0, :math:`m`). All input variables are +tensors with same shape [:math:`d_0`, :math:`d_1`, ..., :math:`d_R`]. +Please note that rank of the input tensor should be at least 2. Each input +variable will be treated as a 2-D matrix with shape [:math:`M`, :math:`N`] +where :math:`M` for :math:`d_0` and :math:`N` for :math:`d_1` * :math:`d_2` +* ... * :math:`d_R`. Let :math:`I_i[j]` be the j-th row of the i-th input +variable. The given index variable should be a 2-D tensor with shape +[:math:`M`, 1]. Let `ID[i]` be the i-th index value of the index variable. +Then the output variable will be a tensor with shape [:math:`d_0`, +:math:`d_1`, ..., :math:`d_R`]. If we treat the output tensor as a 2-D +matrix with shape [:math:`M`, :math:`N`] and let :math:`O[i]` be the i-th +row of the matrix, then `O[i]` is equal to :math:`I_{ID[i]}[i]`. + +* Ids: the index tensor. + +* X[0 : N - 1]: the candidate tensors for output (N >= 2). + +* For each index i from 0 to batchSize - 1, the output is the i-th row of the the (Ids[i])-th tensor. For i-th row of the output tensor: -$$y[i] = x_{k}[i]$$ +$$ +y[i] = x_{k}[i] +$$ -where `y` is the output tensor, `x_{k}` is the k-th input tensor, -and `k = Ids[i]`. +where $y$ is the output tensor, $x_{k}$ is the k-th input tensor, +and $k = Ids[i]$. )DOC"); } diff --git a/paddle/fluid/operators/reader/create_recordio_file_reader_op.cc b/paddle/fluid/operators/reader/create_recordio_file_reader_op.cc index 282ec3f36b98e7aa62d71fb04f72721a5464e21c..559827f08494af6730aafa1e67c46a47c21dedf6 100644 --- a/paddle/fluid/operators/reader/create_recordio_file_reader_op.cc +++ b/paddle/fluid/operators/reader/create_recordio_file_reader_op.cc @@ -78,11 +78,15 @@ class CreateRecordIOReaderOp : public framework::OperatorBase { class CreateRecordIOReaderOpMaker : public FileReaderMakerBase { protected: void Apply() override { - AddAttr("filename", "The filename of record io reader"); + AddAttr( + "filename", + "The filename of record file. This file will given to reader."); AddComment(R"DOC( - CreateRecordIOReader Operator +Open a recordio file and return the reader object. The returned reader object +is thread-safe. - Create a reader from a record io file +NOTE: This is a very low-level API. It is used for debugging data file or +training. Please use `open_files` instead of this API for production usage. )DOC"); } }; diff --git a/paddle/fluid/operators/reader/reader_op_registry.cc b/paddle/fluid/operators/reader/reader_op_registry.cc index 612e1f5eca3a4836db1fd167fc6bb63400d20177..e11256a49ffa6adc9410376cc8a71fa017df7e9c 100644 --- a/paddle/fluid/operators/reader/reader_op_registry.cc +++ b/paddle/fluid/operators/reader/reader_op_registry.cc @@ -54,7 +54,7 @@ std::unique_ptr CreateReaderByFileName( } void FileReaderMakerBase::Make() { - AddOutput("Out", "(ReaderHolder) The created random reader.").AsDuplicable(); + AddOutput("Out", "(ReaderHolder): The created random reader.").AsDuplicable(); AddAttr>("shape_concat", "The concat of all data's shapes."); AddAttr>( "ranks", diff --git a/paddle/fluid/operators/row_conv_op.cc b/paddle/fluid/operators/row_conv_op.cc index 20f140f962c3aac364a1239a663d5f340bbeb6b2..10b1b0c899d833d70fa6afe51998fe210899e3c3 100644 --- a/paddle/fluid/operators/row_conv_op.cc +++ b/paddle/fluid/operators/row_conv_op.cc @@ -78,23 +78,23 @@ class RowConvOpMaker : public framework::OpProtoAndCheckerMaker { public: void Make() override { AddInput("X", - "(LoDTensor), the input(X) is a LodTensor, which supports " + "the input(X) is a LodTensor, which supports " "variable time-length input sequences. The underlying tensor " "in this LoDTensor is a matrix with shape (T x N), where T " "is the total time steps in this mini-batch and N is the input " "data dimension."); AddInput("Filter", - "(Tensor), the input(Filter) is a learnable parameter. It " + "the input(Filter) is a learnable parameter. It " "is a 2-D tensor with shape (future_context x N), where, " "future_context is the future context length and N is the data " "dimension."); AddOutput("Out", - "(LoDTensor), the output(Out) is a LodTensor, which supports " + "the output(Out) is a LodTensor, which supports " "variable time-length input sequences. The underlying tensor " "in this LodTensor is a matrix with shape T x N, i.e., the " "same shape as X."); AddComment(R"DOC( -Row-convolution Operator. +:strong:`Row-convolution operator` The row convolution is called lookahead convolution. This operator was introduced in the following paper for DeepSpeech2: @@ -114,9 +114,23 @@ and a filter ($W$) of size $context \times d$, the output sequence is convolved as: $$ -out_{i, :} = \sum_{j=i}^{i + context} in_{j,:} \dot W_{i-j, :} +out_{i, :} = \\sum_{j=i}^{i + context} in_{j,:} \\cdot W_{i-j, :} $$ +In the above equation: + +* $Out_{i}$: The i-th row of output variable with shape [1, D]. + +* $\\tau$: Future context size. + +* $X_{j}$: The j-th row of input variable with shape [1, D]. + +* $W_{i-j}$: The (i-j)-th row of parameters with shape [1, D]. + +More details about row_conv please refer to +the design document +https://github.com/PaddlePaddle/Paddle/issues/2228#issuecomment-303903645 . + )DOC"); } }; diff --git a/paddle/fluid/operators/split_op.cc b/paddle/fluid/operators/split_op.cc index 5e2b2a994534c2fb1e053c067b36651d358b9da8..d661b276bc31bf0c3ab181d706ffdccec89f0632 100644 --- a/paddle/fluid/operators/split_op.cc +++ b/paddle/fluid/operators/split_op.cc @@ -115,4 +115,7 @@ USE_CPU_ONLY_OP(concat); REGISTER_OPERATOR(split, ops::SplitOp, ops::SplitOpMaker, ops::SplitGradMaker); REGISTER_OP_CPU_KERNEL(split, - ops::SplitOpKernel); + ops::SplitOpKernel, + ops::SplitOpKernel, + ops::SplitOpKernel, + ops::SplitOpKernel); diff --git a/paddle/fluid/operators/split_op.cu.cc b/paddle/fluid/operators/split_op.cu.cc index efa378af857a8881f25c76379ba7cf81e64c80bb..18e0904681753aff7f3deac96efb6d62f389a031 100644 --- a/paddle/fluid/operators/split_op.cu.cc +++ b/paddle/fluid/operators/split_op.cu.cc @@ -15,4 +15,7 @@ limitations under the License. */ #include "paddle/fluid/operators/split_op.h" namespace ops = paddle::operators; REGISTER_OP_CUDA_KERNEL( - split, ops::SplitOpKernel); + split, ops::SplitOpKernel, + ops::SplitOpKernel, + ops::SplitOpKernel, + ops::SplitOpKernel); diff --git a/paddle/fluid/operators/uniform_random_op.cc b/paddle/fluid/operators/uniform_random_op.cc index 137ea91caedabc3167146d91b063dbe9e2e2b931..edd1baa4ace4e246190afcd12b0716f1dd38e243 100644 --- a/paddle/fluid/operators/uniform_random_op.cc +++ b/paddle/fluid/operators/uniform_random_op.cc @@ -86,32 +86,24 @@ class UniformRandomOp : public framework::OperatorWithKernel { class UniformRandomOpMaker : public framework::OpProtoAndCheckerMaker { public: void Make() override { - AddOutput("Out", "(Tensor) The output tensor of uniform random op"); + AddOutput("Out", "The output tensor of uniform random op"); AddComment(R"DOC( -Uniform random operator. - This operator initializes a tensor with random values sampled from a -uniform distribution. +uniform distribution. The random result is in set [min, max]. )DOC"); - AddAttr>("shape", - "(vector) The shape of the output tensor"); - AddAttr("min", - "(float, default -1.0) " - "Minimum value of uniform random") + AddAttr>("shape", "The shape of the output tensor"); + AddAttr("min", "Minimum value of uniform random. [default -1.0].") .SetDefault(-1.0f); - AddAttr("max", - "(float, default 1.0) " - "Maximun value of uniform random") + AddAttr("max", "Maximun value of uniform random. [default 1.0].") .SetDefault(1.0f); AddAttr("seed", - "(int, default 0) " "Random seed used for generating samples. " "0 means use a seed generated by the system." "Note that if seed is not 0, this operator will always " - "generate the same random numbers every time.") + "generate the same random numbers every time. [default 0].") .SetDefault(0); - AddAttr("dtype", "(int, default 5(FP32)) Output tensor data type") + AddAttr("dtype", "Output tensor data type. [default 5(FP32)].") .SetDefault(framework::proto::VarType::FP32); } }; diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py index 80e8ff484a4c04df1b41bbca284d7c604962934c..6d386b764c9b64a2e03635b7b75823f4502d0494 100644 --- a/python/paddle/fluid/layers/control_flow.py +++ b/python/paddle/fluid/layers/control_flow.py @@ -20,6 +20,7 @@ from ..framework import Program, Variable, Operator from ..layer_helper import LayerHelper, unique_name from ..initializer import force_init_on_cpu from ops import logical_and, logical_not, logical_or +import numpy __all__ = [ 'split_lod_tensor', @@ -909,37 +910,40 @@ def create_array(dtype): dtype=dtype) -def less_than(x, y, force_cpu=True, cond=None, **ignored): +@templatedoc() +def less_than(x, y, force_cpu=None, cond=None, **ignored): """ - **Less than** + ${comment} - This layer returns the truth value of :math:`x < y` elementwise. + >>> import paddle.fluid as fluid + >>> less = fluid.layers.less_than(x=label, y=limit) Args: - x(Variable): First operand of *less_than* - y(Variable): Second operand of *less_than* - force_cpu(Bool|True): The output data will be on CPU if set true. + x(${x_type}): ${x_comment}. + y(${y_type}): ${y_comment}. + force_cpu(${force_cpu_type}): ${force_cpu_comment}. cond(Variable|None): Optional output variable to store the result of *less_than* Returns: - Variable: The tensor variable storing the output of *less_than*. - - Examples: - .. code-block:: python - - less = fluid.layers.less_than(x=label, y=limit) + ${out_comment}. """ helper = LayerHelper("less_than", **locals()) if cond is None: cond = helper.create_tmp_variable(dtype='bool') cond.stop_gradient = True + attrs = dict() + if force_cpu is not None: + attrs['force_cpu'] = force_cpu + elif force_init_on_cpu(): + attrs['force_cpu'] = force_init_on_cpu() + helper.append_op( type='less_than', inputs={'X': [x], 'Y': [y]}, outputs={'Out': [cond]}, - attrs={'force_cpu': force_cpu or force_init_on_cpu()}) + attrs=attrs) return cond @@ -1004,8 +1008,28 @@ def array_read(array, i): def shrink_memory(x, i, table): """ - This function creates an operator to shrink_rnn_memory using the RankTable + This function creates an operator to shrink rnn memory using the RankTable as mentioned in the input parameter. + + NOTE: This API is very low-level API. It is used by DynamicRNN only. + + Since the Dynamic RNN uses no-padding way to implement RNN. The sequence + will be sorted by order, and the length of valid memory will be shrink after + each time step. + + Args: + x(Variable): The memory object in the previous time step. + i(Variable): The step count variable. A int scalar as LoDTensor. + table(Variable): The RNNRankTable object. + + Returns: + the memory variable after shrink. + + Examples: + + Since this API is very low level API. The example is not provided. + Please reference the implementation of class DynamicRNN for detail + usage. """ helper = LayerHelper('shrink_memory', **locals()) out = helper.create_tmp_variable(dtype=x.dtype) @@ -1209,6 +1233,34 @@ class IfElseBlockGuard(object): class IfElse(object): + """ + if-else control flow. + + Args: + cond (Variable): condition used to compare. + name (str, default None): The name of this layer. + + Examples: + .. code-block:: python + + limit = fluid.layers.fill_constant_batch_size_like( + input=label, dtype='int64', shape=[1], value=5.0) + cond = fluid.layers.less_than(x=label, y=limit) + ie = fluid.layers.IfElse(cond) + with ie.true_block(): + true_image = ie.input(image) + hidden = fluid.layers.fc(input=true_image, size=100, act='tanh') + prob = fluid.layers.fc(input=hidden, size=10, act='softmax') + ie.output(prob) + + with ie.false_block(): + false_image = ie.input(image) + hidden = fluid.layers.fc( + input=false_image, size=200, act='tanh') + prob = fluid.layers.fc(input=hidden, size=10, act='softmax') + ie.output(prob) + prob = ie() + """ OUT_IF_ELSE_BLOCKS = 0 IN_IF_ELSE_TRUE_BLOCKS = 1 IN_IF_ELSE_FALSE_BLOCKS = 2 @@ -1311,6 +1363,38 @@ class IfElse(object): class DynamicRNN(object): + """ + The dynamic RNN can process a batch of sequence data. The length of each + sample sequence can be different. This API automatically process them in + batch. + + The input lod must be set. Please reference `lod_tensor` + + >>> import paddle.fluid as fluid + >>> data = fluid.layers.data(name='sentence', dtype='int64', lod_level=1) + >>> embedding = fluid.layers.embedding(input=data, size=[65535, 32], + >>> is_sparse=True) + >>> + >>> drnn = fluid.layers.DynamicRNN() + >>> with drnn.block(): + >>> word = drnn.step_input(embedding) + >>> prev = drnn.memory(shape=[200]) + >>> hidden = fluid.layers.fc(input=[word, prev], size=200, act='relu') + >>> drnn.update_memory(prev, hidden) # set prev to hidden + >>> drnn.output(hidden) + >>> + >>> # last is the last time step of rnn. It is the encoding result. + >>> last = fluid.layers.sequence_last_step(drnn()) + + The dynamic RNN will unfold sequence into timesteps. Users need to define + how to process each time step during the :code:`with` block. + + The `memory` is used staging data cross time step. The initial value of + memory can be zero or another variable. + + The dynamic RNN can mark multiple variables as its output. Use `drnn()` to + get the output sequence. + """ BEFORE_RNN = 0 IN_RNN = 1 AFTER_RNN = 2 @@ -1333,6 +1417,15 @@ class DynamicRNN(object): self.mem_link = [] def step_input(self, x): + """ + Mark a sequence as a dynamic RNN input. + Args: + x(Variable): The input sequence. + + Returns: + The current timestep in the input sequence. + + """ self._assert_in_rnn_block_("step_input") if not isinstance(x, Variable): raise TypeError( @@ -1376,6 +1469,15 @@ class DynamicRNN(object): return array_read(array=input_array, i=self.step_idx) def static_input(self, x): + """ + Mark a variable as a RNN input. The input will not be scattered into + time steps. + Args: + x(Variable): The input variable. + + Returns: + The input variable that can access in RNN. + """ self._assert_in_rnn_block_("static_input") if not isinstance(x, Variable): raise TypeError( @@ -1397,6 +1499,10 @@ class DynamicRNN(object): @contextlib.contextmanager def block(self): + """ + The block for user to define operators in RNN. See the class docstring + for more details. + """ if self.status != DynamicRNN.BEFORE_RNN: raise ValueError("rnn.block() can only be invoke once") self.step_idx = fill_constant( @@ -1423,6 +1529,9 @@ class DynamicRNN(object): x=each_array, table=self.lod_rank_table)) def __call__(self, *args, **kwargs): + """ + Get the output of RNN. This API should only be invoked after RNN.block() + """ if self.status != DynamicRNN.AFTER_RNN: raise ValueError(("Output of the dynamic RNN can only be visited " "outside the rnn block.")) @@ -1437,6 +1546,70 @@ class DynamicRNN(object): value=0.0, need_reorder=False, dtype='float32'): + """ + Create a memory variable for dynamic rnn. + + If the :code:`init` is not None, :code:`memory` will be initialized by + this variable. The :code:`need_reorder` is used to reorder the memory as + the input variable. It should be set to true when the initialized memory + depends on the input sample. + + For example, + + >>> import paddle.fluid as fluid + >>> sentence = fluid.layers.data( + >>> name='sentence', dtype='float32', shape=[32]) + >>> boot_memory = fluid.layers.data( + >>> name='boot', dtype='float32', shape=[10]) + >>> + >>> drnn = fluid.layers.DynamicRNN() + >>> with drnn.block(): + >>> word = drnn.step_input(sentence) + >>> memory = drnn.memory(init=boot_memory, need_reorder=True) + >>> hidden = fluid.layers.fc( + >>> input=[word, memory], size=10, act='tanh') + >>> drnn.update_memory(ex_mem=memory, new_mem=hidden) + >>> drnn.output(hidden) + >>> rnn_output = drnn() + + + Otherwise, if :code:`shape`, :code:`value`, :code:`dtype` are set, the + :code:`memory` will be initialized by this :code:`value`. + + For example, + + >>> import paddle.fluid as fluid + >>> sentence = fluid.layers.data( + >>> name='sentence', dtype='float32', shape=[32]) + >>> + >>> drnn = fluid.layers.DynamicRNN() + >>> with drnn.block(): + >>> word = drnn.step_input(sentence) + >>> memory = drnn.memory(shape=[10], dtype='float32', value=0) + >>> hidden = fluid.layers.fc( + >>> input=[word, memory], size=10, act='tanh') + >>> drnn.update_memory(ex_mem=memory, new_mem=hidden) + >>> drnn.output(hidden) + >>> rnn_output = drnn() + + + Args: + init(Variable|None): The initialized variable. + + shape(list|tuple): The memory shape. NOTE the shape does not contain + batch_size. + + value(float): the initalized value. + + need_reorder(bool): True if the initialized memory depends on the + input sample. + + dtype(str|numpy.dtype): The data type of the initialized memory. + + Returns: + the memory variable. + + """ self._assert_in_rnn_block_('memory') if init is not None: if not isinstance(init, Variable): @@ -1504,6 +1677,16 @@ class DynamicRNN(object): return self.memory(init=init) def update_memory(self, ex_mem, new_mem): + """ + Update the memory from ex_mem to new_mem. NOTE that the shape and data + type of :code:`ex_mem` and :code:`new_mem` must be same. + Args: + ex_mem(Variable): the memory variable. + new_mem(Variable): the plain variable generated in RNN block. + + Returns: + None + """ self._assert_in_rnn_block_('update_memory') if not isinstance(ex_mem, Variable): raise TypeError("The input arg `ex_mem` of update_memory() must " @@ -1521,6 +1704,15 @@ class DynamicRNN(object): self.mem_link.append((new_mem, mem_array)) def output(self, *outputs): + """ + mark the RNN output variables. + + Args: + outputs: The output variables. + + Returns: + None + """ self._assert_in_rnn_block_('output') parent_block = self._parent_block_() for each in outputs: diff --git a/python/paddle/fluid/layers/detection.py b/python/paddle/fluid/layers/detection.py index 3a83db12fd13651578deeac6b562bac2f1e4e4b6..edf528a5950ee84be4a3e2097cee36cb5ad8c68e 100644 --- a/python/paddle/fluid/layers/detection.py +++ b/python/paddle/fluid/layers/detection.py @@ -210,53 +210,68 @@ def bipartite_match(dist_matrix, dist_threshold=None, name=None): """ - **Bipartite matchint operator** - - This operator is a greedy bipartite matching algorithm, which is used to - obtain the matching with the maximum distance based on the input + This operator implements a greedy bipartite matching algorithm, which is + used to obtain the matching with the maximum distance based on the input distance matrix. For input 2D matrix, the bipartite matching algorithm can - find the matched column for each row, also can find the matched row for - each column. And this operator only calculate matched indices from column - to row. For each instance, the number of matched indices is the number of - of columns of the input ditance matrix. - - There are two outputs to save matched indices and distance. - A simple description, this algothrim matched the best (maximum distance) + find the matched column for each row (matched means the largest distance), + also can find the matched row for each column. And this operator only + calculate matched indices from column to row. For each instance, + the number of matched indices is the column number of the input distance + matrix. + + There are two outputs, matched indices and distance. + A simple description, this algorithm matched the best (maximum distance) row entity to the column entity and the matched indices are not duplicated in each row of ColToRowMatchIndices. If the column entity is not matched any row entity, set -1 in ColToRowMatchIndices. - Please note that the input DistMat can be LoDTensor (with LoD) or Tensor. + NOTE: the input DistMat can be LoDTensor (with LoD) or Tensor. If LoDTensor with LoD, the height of ColToRowMatchIndices is batch size. If Tensor, the height of ColToRowMatchIndices is 1. + NOTE: This API is a very low level API. It is used by :code:`ssd_loss` + layer. Please consider to use :code:`ssd_loss` instead. + Args: dist_matrix(Variable): This input is a 2-D LoDTensor with shape [K, M]. It is pair-wise distance matrix between the entities represented by each row and each column. For example, assumed one entity is A with shape [K], another entity is B with shape [M]. The - dist_matirx[i][j] is the distance between A[i] and B[j]. The bigger - the distance is, the better macthing the pairs are. Please note, - This tensor can contain LoD information to represent a batch of - inputs. One instance of this batch can contain different numbers of - entities. + dist_matrix[i][j] is the distance between A[i] and B[j]. The bigger + the distance is, the better matching the pairs are. + + NOTE: This tensor can contain LoD information to represent a batch + of inputs. One instance of this batch can contain different numbers + of entities. match_type(string|None): The type of matching method, should be - 'bipartite' or 'per_prediction', 'bipartite' by defalut. + 'bipartite' or 'per_prediction'. [default 'bipartite']. dist_threshold(float|None): If `match_type` is 'per_prediction', this threshold is to determine the extra matching bboxes based - on the maximum distance, 0.5 by defalut. + on the maximum distance, 0.5 by default. Returns: - match_indices(Variable): A 2-D Tensor with shape [N, M] in int type. - N is the batch size. If match_indices[i][j] is -1, it - means B[j] does not match any entity in i-th instance. - Otherwise, it means B[j] is matched to row - match_indices[i][j] in i-th instance. The row number of - i-th instance is saved in match_indices[i][j]. - match_distance(Variable): A 2-D Tensor with shape [N, M] in float type. - N is batch size. If match_indices[i][j] is -1, - match_distance[i][j] is also -1.0. Otherwise, assumed - match_distance[i][j] = d, and the row offsets of each instance - are called LoD. Then match_distance[i][j] = dist_matrix[d+LoD[i]][j]. + tuple: a tuple with two elements is returned. The first is + matched_indices, the second is matched_distance. + + The matched_indices is a 2-D Tensor with shape [N, M] in int type. + N is the batch size. If match_indices[i][j] is -1, it + means B[j] does not match any entity in i-th instance. + Otherwise, it means B[j] is matched to row + match_indices[i][j] in i-th instance. The row number of + i-th instance is saved in match_indices[i][j]. + + The matched_distance is a 2-D Tensor with shape [N, M] in float type + . N is batch size. If match_indices[i][j] is -1, + match_distance[i][j] is also -1.0. Otherwise, assumed + match_distance[i][j] = d, and the row offsets of each instance + are called LoD. Then match_distance[i][j] = + dist_matrix[d+LoD[i]][j]. + + Examples: + + >>> x = fluid.layers.data(name='x', shape=[4], dtype='float32') + >>> y = fluid.layers.data(name='y', shape=[4], dtype='float32') + >>> iou = fluid.layers.iou_similarity(x=x, y=y) + >>> matched_indices, matched_dist = fluid.layers.bipartite_match(iou) """ helper = LayerHelper('bipartite_match', **locals()) match_indices = helper.create_tmp_variable(dtype='int32') @@ -364,7 +379,7 @@ def ssd_loss(location, normalize=True, sample_size=None): """ - **Multi-box loss layer for object dection algorithm of SSD** + **Multi-box loss layer for object detection algorithm of SSD** This layer is to compute dection loss for SSD given the location offset predictions, confidence predictions, prior boxes and ground-truth boudding @@ -372,21 +387,35 @@ def ssd_loss(location, is a weighted sum of the localization loss (or regression loss) and confidence loss (or classification loss) by performing the following steps: - 1. Find matched boundding box by bipartite matching algorithm. + 1. Find matched bounding box by bipartite matching algorithm. + 1.1 Compute IOU similarity between ground-truth boxes and prior boxes. + 1.2 Compute matched boundding box by bipartite matching algorithm. + 2. Compute confidence for mining hard examples + 2.1. Get the target label based on matched indices. + 2.2. Compute confidence loss. + 3. Apply hard example mining to get the negative example indices and update the matched indices. + 4. Assign classification and regression targets + 4.1. Encoded bbox according to the prior boxes. + 4.2. Assign regression targets. + 4.3. Assign classification targets. + 5. Compute the overall objective loss. + 5.1 Compute confidence loss. + 5.1 Compute localization loss. + 5.3 Compute the overall weighted loss. Args: @@ -421,39 +450,36 @@ def ssd_loss(location, mining_type (str): The hard example mining type, should be 'hard_example' or 'max_negative', now only support `max_negative`. normalize (bool): Whether to normalize the SSD loss by the total number - of output locations, True by defalut. + of output locations, True by default. sample_size (int): The max sample size of negative box, used only when mining_type is 'hard_example'. Returns: - Variable: The weighted sum of the localization loss and confidence loss, - with shape [N * Np, 1], N and Np are the same as they are - in `location`. + The weighted sum of the localization loss and confidence loss, with \ + shape [N * Np, 1], N and Np are the same as they are in `location`. Raises: - ValueError: If mining_type is 'hard_example', now only support - mining type of `max_negative`. + ValueError: If mining_type is 'hard_example', now only support mining \ + type of `max_negative`. Examples: - .. code-block:: python - - pb = layers.data( - name='prior_box', - shape=[10, 4], - append_batch_size=False, - dtype='float32') - pbv = layers.data( - name='prior_box_var', - shape=[10, 4], - append_batch_size=False, - dtype='float32') - loc = layers.data(name='target_box', shape=[10, 4], dtype='float32') - scores = layers.data(name='scores', shape=[10, 21], dtype='float32') - gt_box = layers.data( - name='gt_box', shape=[4], lod_level=1, dtype='float32') - gt_label = layers.data( - name='gt_label', shape=[1], lod_level=1, dtype='float32') - loss = layers.ssd_loss(loc, scores, gt_box, gt_label, pb, pbv) + >>> pb = fluid.layers.data( + >>> name='prior_box', + >>> shape=[10, 4], + >>> append_batch_size=False, + >>> dtype='float32') + >>> pbv = fluid.layers.data( + >>> name='prior_box_var', + >>> shape=[10, 4], + >>> append_batch_size=False, + >>> dtype='float32') + >>> loc = fluid.layers.data(name='target_box', shape=[10, 4], dtype='float32') + >>> scores = fluid.layers.data(name='scores', shape=[10, 21], dtype='float32') + >>> gt_box = fluid.layers.data( + >>> name='gt_box', shape=[4], lod_level=1, dtype='float32') + >>> gt_label = fluid.layers.data( + >>> name='gt_label', shape=[1], lod_level=1, dtype='float32') + >>> loss = fluid.layers.ssd_loss(loc, scores, gt_box, gt_label, pb, pbv) """ helper = LayerHelper('ssd_loss', **locals()) diff --git a/python/paddle/fluid/layers/io.py b/python/paddle/fluid/layers/io.py index 9de88e2c3205ace74beff43df7ae8956897d965a..c018664c1e3d326e956a2be9ffe0f0a2c5d7c757 100644 --- a/python/paddle/fluid/layers/io.py +++ b/python/paddle/fluid/layers/io.py @@ -292,6 +292,7 @@ def _copy_reader_create_op_(block, op): return new_op +@templatedoc(op_type='create_recordio_file_reader') def open_recordio_file(filename, shapes, lod_levels, @@ -299,34 +300,30 @@ def open_recordio_file(filename, pass_num=1, for_parallel=True): """ - Open a RecordIO file - - This layer takes a RecordIO file to read from and returns a Reader Variable. - Via the Reader Variable, we can get data from the given RecordIO file. + ${comment} Args: - filename(str): The RecordIO file's name. + filename(${filename_type}): ${filename_comment}. shapes(list): List of tuples which declaring data shapes. - lod_levels(list): List of ints which declaring data lod_level. + lod_levels(${lod_levels_type}): ${lod_levels_comment}. dtypes(list): List of strs which declaring data type. pass_num(int): Number of passes to run. for_parallel(Bool): Set it as True if you are going to run subsequent operators in parallel. Returns: - Variable: A Reader Variable via which we can get RecordIO file data. + ${out_comment}. Examples: - .. code-block:: python - reader = fluid.layers.io.open_recordio_file( - filename='./data.recordio', - shapes=[(3,224,224), (1)], - lod_levels=[0, 0], - dtypes=['float32', 'int64']) - - # Via the reader, we can use 'read_file' layer to get data: - image, label = fluid.layers.io.read_file(reader) + >>> import paddle.fluid as fluid + >>> reader = fluid.layers.io.open_recordio_file( + >>> filename='./data.recordio', + >>> shapes=[(3,224,224), (1)], + >>> lod_levels=[0, 0], + >>> dtypes=['float32', 'int64']) + >>> # Via the reader, we can use 'read_file' layer to get data: + >>> image, label = fluid.layers.io.read_file(reader) """ dtypes = [convert_np_dtype_to_dtype_(dt) for dt in dtypes] shape_concat = [] @@ -554,6 +551,29 @@ def batch(reader, batch_size): def double_buffer(reader, place=None, name=None): + """ + Wrap a double buffer reader. The data will copy to target place with a + double buffer queue. If the target place is None, the place that executor + perform on will be used. + + Args: + reader(Variable): the reader variable need to be wrapped. + place(Place): the place of target data. Default is the sample place of + executor perform. + + name(str): Variable name. None if the user does not care. + + Returns: + wrapped reader with double buffer. + + Examples: + + >>> reader = fluid.layers.open_files(filenames=['somefile'], + >>> shapes=[[-1, 784], [-1, 1]], + >>> dtypes=['float32', 'int64']) + >>> reader = fluid.layers.double_buffer(reader) + >>> img, label = fluid.layers.read_file(reader) + """ attrs = dict() if place is not None: attrs['place'] = str(place).upper() @@ -587,6 +607,26 @@ def read_file(file_obj): class Preprocessor(object): + """ + A block for data pre-processing in reader. + + Args: + reader (Variable): A reader variable. + name (str, default None): The name of the reader. + + Examples: + .. code-block:: python + + preprocessor = fluid.layers.io.Preprocessor(reader=reader) + with preprocessor.block(): + img, lbl = preprocessor.inputs() + img_out = img / 2 + lbl_out = lbl + 1 + preprocessor.outputs(img_out, lbl_out) + + data_file = fluid.layers.io.double_buffer(preprocessor()) + + """ BEFORE_SUB_BLOCK = 0 IN_SUB_BLOCK = 1 AFTER_SUB_BLOCK = 2 diff --git a/python/paddle/fluid/layers/nn.py b/python/paddle/fluid/layers/nn.py index f4a8c5f37d139e1160a6ed7ed37ca6ef7c30e5b0..27fbb0f053e592d156221bb81a56ba3e9c4efe32 100644 --- a/python/paddle/fluid/layers/nn.py +++ b/python/paddle/fluid/layers/nn.py @@ -25,20 +25,72 @@ import utils import random __all__ = [ - 'fc', 'embedding', 'dynamic_lstm', 'dynamic_lstmp', 'dynamic_gru', - 'gru_unit', 'linear_chain_crf', 'crf_decoding', 'cos_sim', 'cross_entropy', - 'square_error_cost', 'chunk_eval', 'sequence_conv', 'conv2d', - 'sequence_pool', 'sequence_softmax', 'softmax', 'pool2d', 'batch_norm', - 'beam_search_decode', 'conv2d_transpose', 'sequence_expand', 'lstm_unit', - 'reduce_sum', 'reduce_mean', 'reduce_max', 'reduce_min', 'reduce_prod', - 'sequence_first_step', 'sequence_last_step', 'dropout', 'split', - 'ctc_greedy_decoder', 'edit_distance', 'l2_normalize', 'matmul', 'topk', - 'warpctc', 'sequence_reshape', 'transpose', 'im2sequence', 'nce', - 'beam_search', 'row_conv', 'multiplex', 'layer_norm', - 'softmax_with_cross_entropy', 'smooth_l1', 'one_hot', - 'autoincreased_step_counter', 'reshape', 'lod_reset', 'lrn', 'pad', - 'label_smooth', 'roi_pool', 'dice_loss', 'image_resize', - 'image_resize_short', 'resize_bilinear', 'gather', 'random_crop', 'mean_iou' + 'fc', + 'embedding', + 'dynamic_lstm', + 'dynamic_lstmp', + 'dynamic_gru', + 'gru_unit', + 'linear_chain_crf', + 'crf_decoding', + 'cos_sim', + 'cross_entropy', + 'square_error_cost', + 'chunk_eval', + 'sequence_conv', + 'conv2d', + 'conv3d', + 'sequence_pool', + 'sequence_softmax', + 'softmax', + 'pool2d', + 'pool3d', + 'batch_norm', + 'beam_search_decode', + 'conv2d_transpose', + 'conv3d_transpose', + 'sequence_expand', + 'lstm_unit', + 'reduce_sum', + 'reduce_mean', + 'reduce_max', + 'reduce_min', + 'reduce_prod', + 'sequence_first_step', + 'sequence_last_step', + 'dropout', + 'split', + 'ctc_greedy_decoder', + 'edit_distance', + 'l2_normalize', + 'matmul', + 'topk', + 'warpctc', + 'sequence_reshape', + 'transpose', + 'im2sequence', + 'nce', + 'beam_search', + 'row_conv', + 'multiplex', + 'layer_norm', + 'softmax_with_cross_entropy', + 'smooth_l1', + 'one_hot', + 'autoincreased_step_counter', + 'reshape', + 'lod_reset', + 'lrn', + 'pad', + 'label_smooth', + 'roi_pool', + 'dice_loss', + 'image_resize', + 'image_resize_short', + 'resize_bilinear', + 'gather', + 'random_crop', + 'mean_iou', ] @@ -1275,8 +1327,6 @@ def conv2d(input, conv2d = fluid.layers.conv2d( input=data, num_filters=2, filter_size=3, act="relu") """ - if stride is None: - stride = [1, 1] num_channels = input.shape[1] @@ -1339,6 +1389,171 @@ def conv2d(input, return helper.append_activation(pre_act) +def conv3d(input, + num_filters, + filter_size, + stride=1, + padding=0, + dilation=1, + groups=None, + param_attr=None, + bias_attr=None, + use_cudnn=True, + use_mkldnn=False, + act=None, + name=None): + """ + **Convlution3D Layer** + + The convolution3D layer calculates the output based on the input, filter + and strides, paddings, dilations, groups parameters. Input(Input) and + Output(Output) are in NCDHW format. Where N is batch size C is the number of + channels, D is the depth of the feature, H is the height of the feature, + and W is the width of the feature. Convlution3D is similar with Convlution2D + but adds one dimension(depth). If bias attribution and activation type are + provided, bias is added to the output of the convolution, and the + corresponding activation function is applied to the final result. + + For each input :math:`X`, the equation is: + + .. math:: + + Out = \sigma (W \\ast X + b) + + In the above equation: + + * :math:`X`: Input value, a tensor with NCDHW format. + * :math:`W`: Filter value, a tensor with MCDHW format. + * :math:`\\ast`: Convolution operation. + * :math:`b`: Bias value, a 2-D tensor with shape [M, 1]. + * :math:`\\sigma`: Activation function. + * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be + different. + + Example: + + - Input: + + Input shape: :math:`(N, C_{in}, D_{in}, H_{in}, W_{in})` + + Filter shape: :math:`(C_{out}, C_{in}, D_f, H_f, W_f)` + + - Output: + Output shape: :math:`(N, C_{out}, D_{out}, H_{out}, W_{out})` + + Where + + .. math:: + + D_{out}&= \\frac{(D_{in} + 2 * paddings[0] - (dilations[0] * (D_f - 1) + 1))}{strides[0]} + 1 \\\\ + H_{out}&= \\frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{strides[1]} + 1 \\\\ + W_{out}&= \\frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{strides[2]} + 1 + + Args: + input (Variable): The input image with [N, C, D, H, W] format. + num_filters(int): The number of filter. It is as same as the output + image channel. + filter_size (int|tuple|None): The filter size. If filter_size is a tuple, + it must contain three integers, (filter_size_D, filter_size_H, filter_size_W). + Otherwise, the filter will be a square. + stride (int|tuple): The stride size. If stride is a tuple, it must + contain three integers, (stride_D, stride_H, stride_W). Otherwise, the + stride_D = stride_H = stride_W = stride. Default: stride = 1. + padding (int|tuple): The padding size. If padding is a tuple, it must + contain three integers, (padding_D, padding_H, padding_W). Otherwise, the + padding_D = padding_H = padding_W = padding. Default: padding = 0. + dilation (int|tuple): The dilation size. If dilation is a tuple, it must + contain three integers, (dilation_D, dilation_H, dilation_W). Otherwise, the + dilation_D = dilation_H = dilation_W = dilation. Default: dilation = 1. + groups (int): The groups number of the Conv3d Layer. According to grouped + convolution in Alex Krizhevsky's Deep CNN paper: when group=2, + the first half of the filters is only connected to the first half + of the input channels, while the second half of the filters is only + connected to the second half of the input channels. Default: groups=1 + param_attr (ParamAttr): The parameters to the Conv3d Layer. Default: None + bias_attr (ParamAttr): Bias parameter for the Conv3d layer. Default: None + use_cudnn (bool): Use cudnn kernel or not, it is valid only when the cudnn + library is installed. Default: True + use_mkldnn (bool): Use mkldnn kernels or not. + act (str): Activation type. Default: None + name (str|None): A name for this layer(optional). If set None, the layer + will be named automatically. + + Returns: + Variable: The tensor variable storing the convolution and \ + non-linearity activation result. + + Raises: + ValueError: If the shapes of input, filter_size, stride, padding and + groups mismatch. + + Examples: + .. code-block:: python + + data = fluid.layers.data( + name='data', shape=[3, 12, 32, 32], dtype='float32') + conv2d = fluid.layers.conv3d( + input=data, num_filters=2, filter_size=3, act="relu") + """ + + l_type = 'conv3d' + + helper = LayerHelper(l_type, **locals()) + dtype = helper.input_dtype() + + num_channels = input.shape[1] + + if groups is None: + num_filter_channels = num_channels + else: + if num_channels % groups != 0: + raise ValueError("num_channels must be divisible by groups.") + num_filter_channels = num_channels / groups + + filter_size = utils.convert_to_list(filter_size, 3, 'filter_size') + stride = utils.convert_to_list(stride, 3, 'stride') + padding = utils.convert_to_list(padding, 3, 'padding') + dilation = utils.convert_to_list(dilation, 3, 'dilation') + + if not isinstance(use_cudnn, bool): + raise ValueError("use_cudnn should be True or False") + + input_shape = input.shape + filter_shape = [num_filters, num_filter_channels] + filter_size + + def _get_default_param_initializer(): + std = (2.0 / (filter_size[0]**3 * num_channels))**0.5 + return Normal(0.0, std, 0) + + filter_param = helper.create_parameter( + attr=helper.param_attr, + shape=filter_shape, + dtype=dtype, + default_initializer=_get_default_param_initializer()) + + pre_bias = helper.create_tmp_variable(dtype) + + helper.append_op( + type=l_type, + inputs={ + 'Input': input, + 'Filter': filter_param, + }, + outputs={"Output": pre_bias}, + attrs={ + 'strides': stride, + 'paddings': padding, + 'dilations': dilation, + 'groups': groups, + 'use_cudnn': use_cudnn, + 'use_mkldnn': use_mkldnn + }) + + pre_act = helper.append_bias_op(pre_bias, dim_start=1, dim_end=2) + + return helper.append_activation(pre_act) + + def sequence_pool(input, pool_type): """ This function add the operator for sequence pooling. @@ -1526,12 +1741,84 @@ def pool2d(input, if not isinstance(use_cudnn, bool): raise ValueError("use_cudnn should be True or False") - helper = LayerHelper('pool2d', **locals()) + l_type = 'pool2d' + + helper = LayerHelper(l_type, **locals()) dtype = helper.input_dtype() pool_out = helper.create_tmp_variable(dtype) helper.append_op( - type="pool2d", + type=l_type, + inputs={"X": input}, + outputs={"Out": pool_out}, + attrs={ + "pooling_type": pool_type, + "ksize": pool_size, + "global_pooling": global_pooling, + "strides": pool_stride, + "paddings": pool_padding, + "use_cudnn": use_cudnn, + "ceil_mode": ceil_mode, + "use_mkldnn": use_mkldnn + }) + + return pool_out + + +def pool3d(input, + pool_size=-1, + pool_type="max", + pool_stride=1, + pool_padding=0, + global_pooling=False, + use_cudnn=True, + ceil_mode=False, + use_mkldnn=False, + name=None): + """ + This function adds the operator for pooling in 3-dimensions, using the + pooling configurations mentioned in input parameters. + + Args: + input (Variable): ${input_comment} + pool_size (int): ${ksize_comment} + pool_type (str): ${pooling_type_comment} + pool_stride (int): stride of the pooling layer. + pool_padding (int): padding size. + global_pooling (bool): ${global_pooling_comment} + use_cudnn (bool): ${use_cudnn_comment} + ceil_mode (bool): ${ceil_mode_comment} + use_mkldnn (bool): ${use_mkldnn_comment} + name (str): A name for this layer(optional). If set None, the layer + will be named automatically. + + Returns: + Variable: output of pool3d layer. + """ + if pool_type not in ["max", "avg"]: + raise ValueError( + "Unknown pool_type: '%s'. It can only be 'max' or 'avg'.", + str(pool_type)) + + if global_pooling is False and pool_size == -1: + raise ValueError( + "When the global_pooling is False, pool_size must be passed " + "and be a valid value. Received pool_size: " + str(pool_size)) + + pool_size = utils.convert_to_list(pool_size, 3, 'pool_size') + pool_padding = utils.convert_to_list(pool_padding, 3, 'pool_padding') + pool_stride = utils.convert_to_list(pool_stride, 3, 'pool_stride') + + if not isinstance(use_cudnn, bool): + raise ValueError("use_cudnn should be True or False") + + l_type = "pool3d" + helper = LayerHelper(l_type, **locals()) + dtype = helper.input_dtype() + pool_out = helper.create_tmp_variable(dtype) + + helper.append_op( + type=l_type, inputs={"X": input}, outputs={"Out": pool_out}, attrs={ @@ -1665,6 +1952,7 @@ def batch_norm(input, return helper.append_activation(batch_norm_out) +@templatedoc() def layer_norm(input, scale=True, shift=True, @@ -1675,20 +1963,11 @@ def layer_norm(input, act=None, name=None): """ - **Layer Normalization** - - Assume feature vectors exist on dimensions - :attr:`begin_norm_axis ... rank(input)` and calculate the moment statistics - along these dimensions for each feature vector :math:`a` with size - :math:`H`, then normalize each feature vector using the corresponding - statistics. After that, apply learnable gain and bias on the normalized - tensor to scale and shift if :attr:`scale` and :attr:`shift` are set. - - Refer to `Layer Normalization `_ + ${comment} The formula is as follows: - .. math:: + .. math:: \\mu & = \\frac{1}{H}\\sum_{i=1}^{H} a_i @@ -1696,6 +1975,15 @@ def layer_norm(input, h & = f(\\frac{g}{\\sigma}(a - \\mu) + b) + * :math:`a`: the vector representation of the summed inputs to the neurons + in that layer. + + * :math:`H`: the number of hidden units in a layers + + * :math:`g`: the trainable scale parameter. + + * :math:`b`: the trainable bias parameter. + Args: input(Variable): The input tensor variable. scale(bool): Whether to learn the adaptive gain :math:`g` after @@ -1714,14 +2002,13 @@ def layer_norm(input, name (str): The name of this layer. It is optional. Returns: - Variable: A tensor variable with the same shape as the input. + ${y_comment} Examples: - .. code-block:: python - data = fluid.layers.data( - name='data', shape=[3, 32, 32], dtype='float32') - x = fluid.layers.layer_norm(input=data, begin_norm_axis=1) + >>> data = fluid.layers.data(name='data', shape=[3, 32, 32], + >>> dtype='float32') + >>> x = fluid.layers.layer_norm(input=data, begin_norm_axis=1) """ helper = LayerHelper('layer_norm', **locals()) dtype = helper.input_dtype() @@ -1952,6 +2239,173 @@ def conv2d_transpose(input, return out +def conv3d_transpose(input, + num_filters, + output_size=None, + filter_size=None, + padding=0, + stride=1, + dilation=1, + groups=None, + param_attr=None, + bias_attr=None, + use_cudnn=True, + act=None, + name=None): + """ + **Convlution3D transpose layer** + + The convolution3D transpose layer calculates the output based on the input, + filter, and dilations, strides, paddings. Input(Input) and output(Output) + are in NCDHW format. Where N is batch size, C is the number of channels, + D is the depth of the feature, H is the height of the feature, and W + is the width of the feature. Parameters(dilations, strides, paddings) are + two elements. These two elements represent height and width, respectively. + The details of convolution transpose layer, please refer to the following + explanation and references `therein `_. + + For each input :math:`X`, the equation is: + + .. math:: + + Out = W \\ast X + + In the above equation: + + * :math:`X`: Input value, a tensor with NCDHW format. + * :math:`W`: Filter value, a tensor with MCDHW format. + * :math:`\\ast` : Convolution transpose operation. + * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be + different. + + Example: + + - Input: + + Input shape: $(N, C_{in}, D_{in}, H_{in}, W_{in})$ + + Filter shape: $(C_{in}, C_{out}, D_f, H_f, W_f)$ + + - Output: + + Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$ + + Where + + .. math:: + + D_{out} &= (D_{in} - 1) * strides[0] - 2 * paddings[0] + dilations[0] * (D_f - 1) + 1 \\\\ + H_{out} &= (H_{in} - 1) * strides[1] - 2 * paddings[1] + dilations[1] * (H_f - 1) + 1 \\\\ + W_{out} &= (W_{in} - 1) * strides[2] - 2 * paddings[2] + dilations[2] * (W_f - 1) + 1 + + Args: + input(Variable): The input image with [N, C, D, H, W] format. + num_filters(int): The number of the filter. It is as same as the output + image channel. + output_size(int|tuple|None): The output image size. If output size is a + tuple, it must contain three integers, (image_D, image_H, image_W). This + parameter only works when filter_size is None. + filter_size(int|tuple|None): The filter size. If filter_size is a tuple, + it must contain three integers, (filter_size_D, filter_size_H, filter_size_W). + Otherwise, the filter will be a square. None if use output size to + calculate filter_size. + padding(int|tuple): The padding size. If padding is a tuple, it must + contain three integers, (padding_D, padding_H, padding_W). Otherwise, the + padding_D = padding_H = padding_W = padding. Default: padding = 0. + stride(int|tuple): The stride size. If stride is a tuple, it must + contain three integers, (stride_D, stride_H, stride_W). Otherwise, the + stride_D = stride_H = stride_W = stride. Default: stride = 1. + dilation(int|tuple): The dilation size. If dilation is a tuple, it must + contain three integers, (dilation_D, dilation_H, dilation_W). Otherwise, the + dilation_D = dilation_H = dilation_W = dilation. Default: dilation = 1. + groups(int): The groups number of the Conv3d transpose layer. Inspired by + grouped convolution in Alex Krizhevsky's Deep CNN paper, in which + when group=2, the first half of the filters is only connected to the + first half of the input channels, while the second half of the + filters is only connected to the second half of the input channels. + Default: groups=1 + param_attr(ParamAttr): The parameters to the Conv3d_transpose Layer. + Default: None + bias_attr(ParamAttr): Bias parameter for the Conv3d layer. Default: None + use_cudnn(bool): Use cudnn kernel or not, it is valid only when the cudnn + library is installed. Default: True + act(str): Activation type. Default: None + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. + + Returns: + Variable: The tensor variable storing the convolution transpose result. + + Raises: + ValueError: If the shapes of input, filter_size, stride, padding and + groups mismatch. + + Examples: + .. code-block:: python + + data = fluid.layers.data( + name='data', shape=[3, 12, 32, 32], dtype='float32') + conv2d_transpose = fluid.layers.conv3d_transpose( + input=data, num_filters=2, filter_size=3) + """ + l_type = "conv3d_transpose" + helper = LayerHelper(l_type, **locals()) + if not isinstance(input, Variable): + raise TypeError("Input of conv3d_transpose must be Variable") + input_channel = input.shape[1] + + padding = utils.convert_to_list(padding, 3, 'padding') + stride = utils.convert_to_list(stride, 3, 'stride') + dilation = utils.convert_to_list(dilation, 3, 'dilation') + + if not isinstance(use_cudnn, bool): + raise ValueError("use_cudnn should be True or False") + + if filter_size is None: + if output_size is None: + raise ValueError("output_size must be set when filter_size is None") + if isinstance(output_size, int): + output_size = [output_size, output_size] + + d_in = input.shape[2] + h_in = input.shape[3] + w_in = input.shape[4] + + filter_size_d = (output_size[0] - (d_in - 1) * stride[0] + 2 * + padding[0] - 1) / dilation[0] + 1 + filter_size_h = (output_size[1] - (h_in - 1) * stride[1] + 2 * + padding[1] - 1) / dilation[1] + 1 + filter_size_w = (output_size[2] - (w_in - 1) * stride[2] + 2 * + padding[2] - 1) / dilation[2] + 1 + filter_size = [filter_size_d, filter_size_h, filter_size_w] + else: + filter_size = utils.convert_to_list(filter_size, 3, + 'conv3d_transpose.filter_size') + + groups = 1 if groups is None else groups + filter_shape = [input_channel, num_filters / groups] + filter_size + img_filter = helper.create_parameter( + dtype=input.dtype, shape=filter_shape, attr=helper.param_attr) + + pre_bias = helper.create_tmp_variable(dtype=input.dtype) + helper.append_op( + type=l_type, + inputs={'Input': [input], + 'Filter': [img_filter]}, + outputs={'Output': pre_bias}, + attrs={ + 'strides': stride, + 'paddings': padding, + 'dilations': dilation, + 'groups': groups, + 'use_cudnn': use_cudnn + }) + + pre_act = helper.append_bias_op(pre_bias, dim_start=1, dim_end=2) + out = helper.append_activation(pre_act) + return out + + def sequence_expand(x, y, ref_level=-1, name=None): """Sequence Expand Layer. This layer will expand the input variable **x** according to specified level lod of **y**. Please note that lod level of @@ -3238,29 +3692,13 @@ def im2sequence(input, filter_size=1, stride=1, padding=0, name=None): return out +@templatedoc() def row_conv(input, future_context_size, param_attr=None, act=None): - """Row Conv Operator. This layer will apply lookahead convolution to - **input**. The input variable should be a 2D LoDTensor with shape [T, D]. - Parameters with shape [future_context_size + 1, D] will be created. The math - equation of row convolution is as follows: - - .. math:: - Out_{i} = \sum_{j = i} ^ {i + \\tau} X_{j} \odot W_{i - j} - - In the above equation: - - * :math:`Out_{i}`: The i-th row of output variable with shape [1, D]. - * :math:`\\tau`: Future context size. - * :math:`X_{j}`: The j-th row of input variable with shape [1, D]. - * :math:`W_{i-j}`: The (i-j)-th row of parameters with shape [1, D]. - - More details about row_conv please refer to the paper \ - (http://www.cs.cmu.edu/~dyogatam/papers/wang+etal.iclrworkshop2016.pdf) and - the design document \ - (https://github.com/PaddlePaddle/Paddle/issues/2228#issuecomment-303903645). + """ + ${comment} Args: - input (Variable): Input variable, a 2D LoDTensor with shape [T, D]. + input (${x_type}): ${x_comment}. future_context_size (int): Future context size. Please note, the shape of convolution kernel is [future_context_size + 1, D]. param_attr (ParamAttr): Attributes of parameters, including @@ -3268,14 +3706,13 @@ def row_conv(input, future_context_size, param_attr=None, act=None): act (str): Non-linear activation to be applied to output variable. Returns: - Variable: The output tensor with same shape as input tensor. + ${out_comment}. Examples: - .. code-block:: python - - x = fluid.layers.data(name='x', shape=[16], - dtype='float32', lod_level=1) - out = fluid.layers.row_conv(input=x, future_context_size=2) + >>> import paddle.fluid as fluid + >>> x = fluid.layers.data(name='x', shape=[16], + >>> dtype='float32', lod_level=1) + >>> out = fluid.layers.row_conv(input=x, future_context_size=2) """ helper = LayerHelper('row_conv', **locals()) dtype = helper.input_dtype() @@ -3291,42 +3728,23 @@ def row_conv(input, future_context_size, param_attr=None, act=None): return helper.append_activation(out) +@templatedoc() def multiplex(inputs, index): """ - **Multiplex Layer** - - Referring to the given index variable, this layer selects rows from the - input variables to construct a multiplex variable. Assuming that there are - :math:`m` input variables and :math:`I_i` represents the i-th input - variable and :math:`i` is in [0, :math:`m`). All input variables are - tensors with same shape [:math:`d_0`, :math:`d_1`, ..., :math:`d_R`]. - Please note that rank of the input tensor should be at least 2. Each input - variable will be treated as a 2-D matrix with shape [:math:`M`, :math:`N`] - where :math:`M` for :math:`d_0` and :math:`N` for :math:`d_1` * :math:`d_2` - * ... * :math:`d_R`. Let :math:`I_i[j]` be the j-th row of the i-th input - variable. The given index variable should be a 2-D tensor with shape - [:math:`M`, 1]. Let `ID[i]` be the i-th index value of the index variable. - Then the output variable will be a tensor with shape [:math:`d_0`, - :math:`d_1`, ..., :math:`d_R`]. If we treat the output tensor as a 2-D - matrix with shape [:math:`M`, :math:`N`] and let :math:`O[i]` be the i-th - row of the matrix, then `O[i]` is equal to :math:`I_{ID[i]}[i]`. + ${comment} + + >>> import paddle.fluid as fluid + >>> x1 = fluid.layers.data(name='x1', shape=[4], dtype='float32') + >>> x2 = fluid.layers.data(name='x2', shape=[4], dtype='float32') + >>> index = fluid.layers.data(name='index', shape=[1], dtype='int32') + >>> out = fluid.layers.multiplex(inputs=[x1, x2], index=index) Args: - inputs (list): A list of variables to gather from. All variables have the - same shape and the rank is at least 2. - index (Variable): Tensor, index variable which is a 2-D tensor - with shape [M, 1] where M is the batch size. + inputs (list): ${x_comment}. + index (${ids_type}): ${ids_comment}. Returns: - Variable: Multiplex variable gathered from input variables. - - Examples: - .. code-block:: python - - x1 = fluid.layers.data(name='x1', shape=[4], dtype='float32') - x2 = fluid.layers.data(name='x2', shape=[4], dtype='float32') - index = fluid.layers.data(name='index', shape=[1], dtype='int32') - out = fluid.layers.multiplex(inputs=[x1, x2], index=index) + ${out_comment}. """ helper = LayerHelper('multiplex', **locals()) diff --git a/python/paddle/fluid/layers/ops.py b/python/paddle/fluid/layers/ops.py index 98f169e8f0881fbba6aecb45b43a52c8fd51132d..6f404c5cc608abda91c1d042d405f109dedc55c9 100644 --- a/python/paddle/fluid/layers/ops.py +++ b/python/paddle/fluid/layers/ops.py @@ -40,8 +40,6 @@ __activations__ = [ 'relu6', 'pow', 'stanh', - 'hard_shrink', - 'thresholded_relu', 'hard_sigmoid', 'swish', ] @@ -64,11 +62,9 @@ __all__ = [ 'logical_or', 'logical_xor', 'logical_not', - 'uniform_random', 'uniform_random_batch_size_like', 'gaussian_random', 'gaussian_random_batch_size_like', - 'cumsum', 'scatter', 'sum', 'slice', @@ -79,3 +75,88 @@ __all__ = [ for _OP in set(__all__): globals()[_OP] = generate_layer_fn(_OP) + +__all__ += ["uniform_random"] + +_uniform_random_ = generate_layer_fn('uniform_random') + + +def uniform_random(shape, dtype=None, min=None, max=None, seed=None): + kwargs = dict() + for name in locals(): + val = locals()[name] + if val is not None: + kwargs[name] = val + return _uniform_random_(**kwargs) + + +uniform_random.__doc__ = _uniform_random_.__doc__ + """ +Examples: + + >>> result = fluid.layers.uniform_random(shape=[32, 784]) +""" + +__all__ += ['hard_shrink'] + +_hard_shrink_ = generate_layer_fn('hard_shrink') + + +def hard_shrink(x, threshold=None): + kwargs = dict() + for name in locals(): + val = locals()[name] + if val is not None: + kwargs[name] = val + return _hard_shrink_(**kwargs) + + +hard_shrink.__doc__ = _hard_shrink_.__doc__ + """ +Examples: + + >>> data = fluid.layers.data(name="input", shape=[784]) + >>> result = fluid.layers.hard_shrink(x=data, threshold=0.3) +""" + +__all__ += ['cumsum'] + +_cum_sum_ = generate_layer_fn('cumsum') + + +def cumsum(x, axis=None, exclusive=None, reverse=None): + kwargs = dict() + for name in locals(): + val = locals()[name] + if val is not None: + kwargs[name] = val + + return _cum_sum_(**kwargs) + + +cumsum.__doc__ = _cum_sum_.__doc__ + """ +Examples: + + >>> data = fluid.layers.data(name="input", shape=[32, 784]) + >>> result = fluid.layers.cumsum(data, axis=0) +""" + +__all__ += ['thresholded_relu'] + +_thresholded_relu_ = generate_layer_fn('thresholded_relu') + + +def thresholded_relu(x, threshold=None): + kwargs = dict() + for name in locals(): + val = locals()[name] + if val is not None: + kwargs[name] = val + + _thresholded_relu_(**kwargs) + + +thresholded_relu.__doc__ = _thresholded_relu_.__doc__ + """ +Examples: + + >>> data = fluid.layers.data(name="input", shape=[1]) + >>> result = fluid.layers.thresholded_relu(data, threshold=0.4) +""" diff --git a/python/paddle/fluid/layers/tensor.py b/python/paddle/fluid/layers/tensor.py index 62b01d595a812ee8fc094e40b6dfb5c3f56cd012..04efc40af5ef36a81ef23f537130d8ca54d73365 100644 --- a/python/paddle/fluid/layers/tensor.py +++ b/python/paddle/fluid/layers/tensor.py @@ -6,7 +6,7 @@ # # http://www.apache.org/licenses/LICENSE-2.0 # -# Unless required by applicable law or agreed to in writing, software +# Unlessf required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and @@ -51,7 +51,12 @@ def create_parameter(shape, is_bias=False, default_initializer=None): """ - Create a parameter + Create a parameter. The parameter is a learnable variable, which can have + gradient, and can be optimized. + + NOTE: this is a very low-level API. This API is useful when you create + operator by your self. instead of using layers. + Args: shape(list[int]): shape of the parameter dtype(string): element type of the parameter @@ -63,7 +68,12 @@ def create_parameter(shape, default_initializer(Initializer): initializer for the parameter Returns: - Parameter: the created parameter + the created parameter. + + Examples: + >>> W = fluid.layers.create_parameter(shape=[784, 200], dtype='float32') + >>> data = fluid.layers.data(name="img", shape=[64, 784], append_batch_size=False) + >>> hidden = fluid.layers.matmul(x=data, y=W) """ helper = LayerHelper("create_parameter", **locals()) if attr is None: diff --git a/python/paddle/fluid/tests/book/test_label_semantic_roles.py b/python/paddle/fluid/tests/book/test_label_semantic_roles.py index bc8a1aafc82d62501cecfa71be0cc3851c75eae2..99d51ae0076178aca50e36c2c187257a8ba1cbf2 100644 --- a/python/paddle/fluid/tests/book/test_label_semantic_roles.py +++ b/python/paddle/fluid/tests/book/test_label_semantic_roles.py @@ -76,8 +76,7 @@ def db_lstm(word, predicate, ctx_n2, ctx_n1, ctx_0, ctx_p1, ctx_p2, mark, emb_layers.append(mark_embedding) hidden_0_layers = [ - fluid.layers.fc(input=emb, size=hidden_dim, act='tanh') - for emb in emb_layers + fluid.layers.fc(input=emb, size=hidden_dim) for emb in emb_layers ] hidden_0 = fluid.layers.sums(input=hidden_0_layers) @@ -94,8 +93,8 @@ def db_lstm(word, predicate, ctx_n2, ctx_n1, ctx_0, ctx_p1, ctx_p2, mark, for i in range(1, depth): mix_hidden = fluid.layers.sums(input=[ - fluid.layers.fc(input=input_tmp[0], size=hidden_dim, act='tanh'), - fluid.layers.fc(input=input_tmp[1], size=hidden_dim, act='tanh') + fluid.layers.fc(input=input_tmp[0], size=hidden_dim), + fluid.layers.fc(input=input_tmp[1], size=hidden_dim) ]) lstm = fluid.layers.dynamic_lstm( diff --git a/python/paddle/fluid/transpiler/memory_optimization_transpiler.py b/python/paddle/fluid/transpiler/memory_optimization_transpiler.py index 9ff0ae6fca27d4681891b2033e2f8f95bd825942..8bfb554845d9b128f000d6c90cf626416a198eef 100644 --- a/python/paddle/fluid/transpiler/memory_optimization_transpiler.py +++ b/python/paddle/fluid/transpiler/memory_optimization_transpiler.py @@ -157,9 +157,11 @@ class ControlFlowGraph(object): if op.type() == "fill_constant" and op.attr("force_cpu") == True: self._skip_opt.update(op.output_arg_names()) - def release_memory(self): + def release_memory(self, skip_opt_set=None): self._dataflow_analyze() self._update_skip_opt_set() + if skip_opt_set: + self._skip_opt.update(skip_opt_set) fwd_id = 0 bwd_id = 0 for i in range(self.op_size): @@ -183,7 +185,7 @@ class ControlFlowGraph(object): else: bwd_id += 1 - def memory_optimize(self, level=0): + def memory_optimize(self, skip_opt_set=None, level=0): def compare_shape(x_shape, cache_shape, opt_level): if opt_level == 0: return x_shape == cache_shape @@ -200,6 +202,9 @@ class ControlFlowGraph(object): self._dataflow_analyze() self._update_skip_opt_set() + # update skip set to meet users' demand + if skip_opt_set: + self._skip_opt.update(skip_opt_set) self.pool = [] for i in range(self.op_size): op = self._ops[i] @@ -358,7 +363,7 @@ def _get_cfgs(input_program): return cfgs -def memory_optimize(input_program, print_log=False, level=0): +def memory_optimize(input_program, skip_opt_set=None, print_log=False, level=0): """Optimize memory by reusing var memory. Note: it doesn't not support subblock nested in subblock. @@ -374,10 +379,10 @@ def memory_optimize(input_program, print_log=False, level=0): PRINT_LOG = print_log cfgs = _get_cfgs(input_program) for cfg in cfgs: - cfg.memory_optimize(level) + cfg.memory_optimize(skip_opt_set=skip_opt_set, level=level) -def release_memory(input_program): +def release_memory(input_program, skip_opt_set=None): cfgs = _get_cfgs(input_program) for cfg in cfgs: - cfg.release_memory() + cfg.release_memory(skip_opt_set=skip_opt_set)