2017 11 29 - PaddlePaddle/Paddle GitHub Wiki

• iOS Camera Object detection SSD Demo https://github.com/PaddlePaddle/Mobile/pull/26
• Android Camera Object detection SSD Demo https://github.com/PaddlePaddle/Mobile/pull/38

• Fix python.v2.fluid arg parse
  • https://github.com/PaddlePaddle/Paddle/pull/6055
• refine test_recognize_digits_mlp and format codes
  • https://github.com/PaddlePaddle/Paddle/pull/5937
• fix bug in infer.py of text classification
  • https://github.com/PaddlePaddle/models/pull/492

I make a statistic of recent jobs on teamcity, 9 jobs failed because of such error " Insufficient GPU memory to allocation" among 101 jobs. That's to say about 10% jobs will fail unexpectedly.

• Refine unittest with setting gflags
  • https://github.com/PaddlePaddle/Paddle/pull/5476

• TFLite survey: https://github.com/PaddlePaddle/Mobile/wiki/Tensorflow-Lite-(Introduction,-motivation-and-few-components-in-depth)
• FlatBuffers survey: https://github.com/PaddlePaddle/Mobile/wiki/FlatBuffers
• Survey for Tensorflow multi-device: https://github.com/PaddlePaddle/Paddle/wiki/Understanding-Multi-device-training-in-TensorFlow
• Wrap up log loss: https://github.com/PaddlePaddle/Paddle/pull/5854
• Fix Latex equations: https://github.com/PaddlePaddle/Paddle/pull/6009
• Working on few other LateX issues: https://github.com/PaddlePaddle/Paddle/issues/6069

• FP16
  • [WIP] float16 data type support: https://github.com/PaddlePaddle/Paddle/pull/5716
  • Investigate half support on different versions of CUDA: https://github.com/PaddlePaddle/Paddle/pull/5851
• Incorrectly displayed doc:
  • https://github.com/PaddlePaddle/Paddle/issues/5973
  • https://github.com/PaddlePaddle/Paddle/issues/6067
  • https://github.com/PaddlePaddle/Paddle/issues/6068
  • https://github.com/PaddlePaddle/Paddle/issues/6069
  • https://github.com/PaddlePaddle/Paddle/issues/6070
• Code review
  • https://github.com/PaddlePaddle/Paddle/pull/6058#pullrequestreview-80066138
  • https://github.com/PaddlePaddle/Paddle/pull/6009#pullrequestreview-79764855
  • https://github.com/PaddlePaddle/Paddle/pull/5749#pullrequestreview-77576491

• Better Implementation of GAN
  • https://github.com/PaddlePaddle/Paddle/pull/5743
• A strong typed ParamAttr
  • https://github.com/PaddlePaddle/Paddle/pull/5996
• A python wrapper for conv_transpose_layer
  • https://github.com/PaddlePaddle/Paddle/pull/5946
• Unify fluid into one package
  • https://github.com/PaddlePaddle/Paddle/pull/5924
• Add Better API for switch program
  • https://github.com/PaddlePaddle/Paddle/pull/5932
• Several Enhancement
  • https://github.com/PaddlePaddle/Paddle/pull/6062
  • https://github.com/PaddlePaddle/Paddle/pull/5978
  • https://github.com/PaddlePaddle/Paddle/pull/5942
  • https://github.com/PaddlePaddle/Paddle/pull/5824
  • https://github.com/PaddlePaddle/Paddle/pull/5873
  • https://github.com/PaddlePaddle/Paddle/pull/6040

• Update conv_seq2seq

  https://github.com/PaddlePaddle/models/pull/493

• Update doc of layers.py

  https://github.com/PaddlePaddle/Paddle/pull/5890

  https://github.com/PaddlePaddle/Paddle/pull/5962

• Discussion and design of cluster training

  • https://github.com/PaddlePaddle/Paddle/pull/5776#discussion_r153366296

• Imperative programming of Paddle

  • https://github.com/PaddlePaddle/Paddle/pull/6076

• Finished DeepFM model [merged]:
  • https://github.com/PaddlePaddle/models/pull/485
  • https://github.com/PaddlePaddle/models/pull/494
• Restructure the code of scheduled sampling [merged]:
  • https://github.com/PaddlePaddle/models/pull/498

• Fix lstm_op and gru_op in debug mode.
  • https://github.com/PaddlePaddle/Paddle/pull/5872
• Benchmark:
  • Profiling tools survey: https://github.com/PaddlePaddle/Paddle/issues/5966
  • Add CUDA profiler tools in the new framework.
    • https://github.com/PaddlePaddle/Paddle/pull/5954
    • https://github.com/PaddlePaddle/Paddle/pull/6022
  • Add benchmark configuration and scripts for convnets.
    • https://github.com/PaddlePaddle/Paddle/pull/6056
• Work about mobile
  • Fix bug in RoI pooling
    • https://github.com/PaddlePaddle/Paddle/pull/5919
  • Cross-compling for android with protobuf-3.2
    • https://github.com/qingqing01/protobuf/pull/1
  • Help OCR to debug text detection after integrating PaddlePaddle lib in their tools.
• Code Review:
  • https://github.com/PaddlePaddle/Paddle/pull/5826
  • https://github.com/PaddlePaddle/Paddle/pull/5837#pullrequestreview-79853637
  • https://github.com/PaddlePaddle/Paddle/pull/6055
  • https://github.com/PaddlePaddle/Paddle/pull/6021
  • https://github.com/PaddlePaddle/Paddle/pull/5946#pullrequestreview-79131656
  • https://github.com/PaddlePaddle/Paddle/pull/5831#pullrequestreview-78875631
  • https://github.com/PaddlePaddle/Paddle/pull/5927

• TeamCity CI, add a project that update latest-dev Docker image nightly:https://github.com/PaddlePaddle/Paddle/issues/6010
• Update Hsigmoid Op PR, add backward implement: https://github.com/PaddlePaddle/Paddle/pull/5676
• Discuss cluster training,https://github.com/PaddlePaddle/Paddle/issues/5958
• Bug Fix:
  • https://github.com/PaddlePaddle/Paddle/pull/5948
  • https://github.com/PaddlePaddle/Paddle/pull/6050
  • https://github.com/PaddlePaddle/Paddle/pull/5906
  • https://github.com/PaddlePaddle/buildtools/pull/6
• Review:
  • Paddle on Cuda9: https://github.com/PaddlePaddle/Paddle/pull/5956

• fix grep socket error in lscpu command: https://github.com/PaddlePaddle/Paddle/pull/5952
• code review:
  • Update installation docs: https://github.com/PaddlePaddle/Paddle/pull/5818
  • Refine install docs: https://github.com/PaddlePaddle/Paddle/pull/5943
  • polish mkldnn doc: https://github.com/PaddlePaddle/Paddle/pull/5910
  • enable inference benchmark: https://github.com/PaddlePaddle/Paddle/pull/5933
• Prepare PPT for Intel Vertical Summit(12.1-12.3)
• Discuss update images for models: https://github.com/PaddlePaddle/models/issues/499

Deep Speech 2:

• Train en8k model with learning rate decay & clipping by global norm & padding removing

• Update Docker image and build Docker image for cloud training

  • https://github.com/PaddlePaddle/DeepSpeech/issues/33
  • https://github.com/PaddlePaddle/DeepSpeech/issues/34

• Fix doc, setup and decoder

  • https://github.com/PaddlePaddle/DeepSpeech/pull/37
  • https://github.com/PaddlePaddle/DeepSpeech/pull/29
  • https://github.com/PaddlePaddle/DeepSpeech/pull/21
  • https://github.com/PaddlePaddle/DeepSpeech/pull/19

Op porting:

• Upate ctc edit distance operator
  • https://github.com/PaddlePaddle/Paddle/pull/5300
• Revise license and comments in reshape_op and rank_loss_op
  • https://github.com/PaddlePaddle/Paddle/pull/6060
  • https://github.com/PaddlePaddle/Paddle/pull/6058

• Fix crop layer and switch order layer.
  • https://github.com/PaddlePaddle/Paddle/pull/5888
• Refine nce op
  • https://github.com/PaddlePaddle/Paddle/pull/5480
• Add math function for sampling integers
  • https://github.com/PaddlePaddle/Paddle/pull/5945
• Implement Policy Gradients demo with refactored API
  • https://github.com/PaddlePaddle/models/pull/505

• Review
  • https://github.com/PaddlePaddle/Paddle/pull/5716
  • https://github.com/PaddlePaddle/Paddle/pull/5976
  • https://github.com/PaddlePaddle/Paddle/issues/6005
  • https://github.com/PaddlePaddle/Paddle/pull/5851
  • https://github.com/PaddlePaddle/Paddle/pull/6051
  • https://github.com/PaddlePaddle/Mobile/pull/29

• PR:
  • Fix the check in addto_layer
    • https://github.com/PaddlePaddle/Paddle/pull/5927
  • Refine roi_pool_op
    • https://github.com/PaddlePaddle/Paddle/pull/5931
  • Fix gru_op related code style
    • https://github.com/PaddlePaddle/Paddle/pull/6021
  • Fix useGpu in HierarchicalSigmoidLayer
    • https://github.com/PaddlePaddle/Paddle/pull/6007
  • Refine NER demo in models
    • https://github.com/PaddlePaddle/models/pull/504
  • Enhance AvgPooling in PoolLayer to support include_mode(WIP)
  • Add configurations for inception-resnet-v2 in models(WIP)
• PR Review:
  • https://github.com/PaddlePaddle/Paddle/pull/5831
  • https://github.com/PaddlePaddle/Paddle/pull/5919
  • https://github.com/PaddlePaddle/models/pull/484
  • https://github.com/PaddlePaddle/Paddle/pull/6062

• Mobile deployment
  • Help ocr to integrate Paddle to their mobile system
  • Enable the case N != ldc in EigenBlasGemm
    • https://github.com/PaddlePaddle/Paddle/pull/5976
  • Add a c-api interface to initialize the thread environment of Paddle
    • https://github.com/PaddlePaddle/Paddle/pull/5773
  • Fix the problem that building for Android fails with WITH_TESTING=ON
    • https://github.com/PaddlePaddle/Paddle/pull/6051
  • Fix the linking error for iOS simulator (architecture x86_64)
    • https://github.com/PaddlePaddle/Paddle/pull/6081

• unpool 2dmax https://github.com/PaddlePaddle/Paddle/issues/5787 done
• spp https://github.com/PaddlePaddle/Paddle/issues/5988 doing coding35%

• SSD detection algorithm:
  • Prior Box Operator (Doing)
    • https://github.com/PaddlePaddle/Paddle/issues/6015

• Enet:

  • It is now consistent with the Caffe model
  • prelu neon impl https://github.com/PaddlePaddle/Paddle/issues/6063

• OpenPose:

  • in process(convert models and speed test).

• TensorFlowLite investigation.
  https://github.com/PaddlePaddle/Mobile/wiki/TfLite-Survey

• pr

  • https://github.com/PaddlePaddle/Mobile/pull/37 fix inference doc of Mobile
  • https://github.com/PaddlePaddle/Paddle/pull/5641 fix prelu api
  • https://github.com/PaddlePaddle/Paddle/pull/5807 fix prelu doc

• Distributed training design milestone discussion: https://github.com/PaddlePaddle/Paddle/issues/5958
• Add ExecutionPlan design https://github.com/PaddlePaddle/Paddle/pull/6078

• https://github.com/PaddlePaddle/Paddle/issues/6074
• https://github.com/PaddlePaddle/Paddle/issues/6071
• https://github.com/PaddlePaddle/Paddle/issues/5866
• https://github.com/PaddlePaddle/Paddle/issues/5865
• https://github.com/PaddlePaddle/Paddle/pull/6073

• Dynamic RNN:
  • https://github.com/PaddlePaddle/Paddle/pull/5950
• related fixes:
  • https://github.com/PaddlePaddle/Paddle/pull/6030
  • https://github.com/PaddlePaddle/Paddle/pull/6038
  • https://github.com/PaddlePaddle/Paddle/pull/5913
  • https://github.com/PaddlePaddle/Paddle/pull/5935
  • https://github.com/PaddlePaddle/Paddle/pull/5986

PR:

• Refine code(momentum_op, rmsprop_op)
  • https://github.com/PaddlePaddle/Paddle/pull/6000
• [Add checking for c-api] outputArg is empty or not.
  • https://github.com/PaddlePaddle/Paddle/pull/5822
• fix conv and conv_trans op doc(merged)
  • https://github.com/PaddlePaddle/Paddle/pull/5951
• remove vector::erase(merged)
  • https://github.com/PaddlePaddle/Paddle/pull/5791
• CPU performance tuning

Review

• Add size, height and width for crop layer and switch order layer.
  • https://github.com/PaddlePaddle/Paddle/pull/5888
• Conv cudnn 3d
  • https://github.com/PaddlePaddle/Paddle/pull/5783
• Feature/cpu profiling
  • https://github.com/PaddlePaddle/Paddle/pull/5895

• feature

  • batch norm support matrix input
  • Make enforce target
  • update gflags
  • py_test and test_image_classification_train support argument

• profile

  • prepare a machine with docker and GPU for profiling

• inference && C-API discussion

• review

  • Feature/param attr

• Refactorization:
  • benchmark and profiler add profile benchmark and tools
    • https://github.com/PaddlePaddle/Paddle/issues/6024 benchmark scripts
    • https://github.com/dzhwinter/benchmark
  • add reduce product
    • https://github.com/PaddlePaddle/Paddle/pull/5903
  • add ceil and floor
    • https://github.com/PaddlePaddle/Paddle/pull/5898
  • fixs
    • https://github.com/PaddlePaddle/Paddle/pull/5455
    • https://github.com/PaddlePaddle/Paddle/pull/5918

• PR:

  • Fix latex syntax in bilinear tensor product operator.

    • https://github.com/PaddlePaddle/Paddle/pull/5875

  • Fix hsigmoid_layer when using GPU.

    • https://github.com/PaddlePaddle/Paddle/pull/5926

  • Add sum eval for generate_sequence_by_rnn_lm.

    • https://github.com/PaddlePaddle/models/pull/502

• Review:

  • Fix useGpu in HierarchicalSigmoidLayer

    • https://github.com/PaddlePaddle/Paddle/pull/6007

• OpTest enhancement: enhance operator unit test base class, rewrite several methods && add feature && bug fix: https://github.com/PaddlePaddle/Paddle/pull/6064
• BatchNorm enhancement support 2D and 5D: add GPU implementation(PR not update yet) https://github.com/PaddlePaddle/Paddle/pull/5845

• Compile and run ds2 on PX2
  https://github.com/PaddlePaddle/Paddle/pull/6027
• Refine Doc for smooth l1 loss op
  https://github.com/PaddlePaddle/Paddle/pull/6031
• Train new network on BaiduEN8k and Unicom Data
• Code Review
  https://github.com/PaddlePaddle/DeepSpeech/pull/37

• Inference API
  • Android NN Survey https://github.com/PaddlePaddle/Mobile/issues/30 https://github.com/PaddlePaddle/Mobile/wiki/Android-NN-Survey
  • CoreML Survey https://github.com/PaddlePaddle/Mobile/issues/35 https://github.com/PaddlePaddle/Mobile/wiki/CoreML-Survey
• Documentation
  • Fix latex comments in sgd_op.cc https://github.com/PaddlePaddle/Paddle/pull/5940
• Code Review
  • https://github.com/PaddlePaddle/Paddle/pull/6073#pullrequestreview-80094730

• Update how to write documentation: https://github.com/PaddlePaddle/Paddle/pull/6014
• Provide an alternative method to activate the viewer tool: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/346
• Prepare for the presentation.
• Update Models images: https://github.com/PaddlePaddle/models/issues/499

• Add operator viewer on paddlepaddle.org doc_mode
  • https://github.com/PaddlePaddle/PaddlePaddle.org/pull/344
• Getting up to speed on Paddleboard
• Prepare project planning documentation for Paddleboard, Paddle on Mobile, and PaddlePaddle.org

• Worked on PaddleBoard proposal through conversations across the team and beyond
• Fixed operators conversion Math issues: https://github.com/PaddlePaddle/PaddlePaddle.org/pull/342
• Studied more deep learning stuff

• Hinge loss operator PR: https://github.com/PaddlePaddle/Paddle/pull/5837
• Row conv operator (still in progress), PR: https://github.com/PaddlePaddle/Paddle/pull/6013