2017 05 17

wangkuiyi

• Refactorization of package math

  • https://github.com/PaddlePaddle/Paddle/pull/2099#issuecomment-300824345
  • https://github.com/PaddlePaddle/Paddle/pull/2091
  • https://github.com/PaddlePaddle/Paddle/pull/2118#pullrequestreview-37894735
  • https://github.com/PaddlePaddle/Paddle/pull/2126#pullrequestreview-37970745
  • https://github.com/PaddlePaddle/Paddle/pull/2149

• Build System

  • https://github.com/PaddlePaddle/Paddle/issues/2154
  • https://github.com/PaddlePaddle/Paddle/pull/2172#pullrequestreview-38540594
  • https://github.com/PaddlePaddle/Paddle/pull/2187
  • https://github.com/PaddlePaddle/Paddle/pull/2180#pullrequestreview-38812873

• Refactorization of package layer

  • https://github.com/PaddlePaddle/Paddle/pull/2104#pullrequestreview-37747930

• PaddlePaddle Cloud

  • https://github.com/PaddlePaddle/Paddle/pull/2115#pullrequestreview-37860575

caoying

• code review
  • https://github.com/PaddlePaddle/models/pull/24 [merged]
  • https://github.com/PaddlePaddle/models/pull/19 [merged]
  • https://github.com/PaddlePaddle/models/pull/43
  • https://github.com/PaddlePaddle/models/pull/29
  • https://github.com/PaddlePaddle/models/pull/22
  • https://github.com/PaddlePaddle/Paddle/pull/2104
• fix a bug of NCE layer
  • https://github.com/PaddlePaddle/Paddle/pull/2134
• update the SRL demo
  • https://github.com/PaddlePaddle/Paddle/pull/2178
  • add the inferring process and an evaluator
  • accelerate the inferring process by following the advice given by yuyang

helinwang

• Paddle Cloud
  • PR: PServer client library: stub and cgo part with functional tests. https://github.com/PaddlePaddle/Paddle/pull/2172
  • PR: Implement PServer RPC Service, gradient update logic, cgo part. https://github.com/PaddlePaddle/Paddle/pull/2188
  • Discussion: https://github.com/PaddlePaddle/Paddle/issues/2169
  • Discussion: https://github.com/PaddlePaddle/Paddle/issues/2168
  • Review: https://github.com/PaddlePaddle/Paddle/pull/2111
  • Review: https://github.com/PaddlePaddle/Paddle/pull/2083
  • Review: https://github.com/PaddlePaddle/cloud/pull/30
• Build System Discussions
  • Review: https://github.com/PaddlePaddle/Paddle/pull/2187
  • Review: https://github.com/PaddlePaddle/Paddle/pull/2182

gongweibao

• PaddlePaddle:

  • PR: https://github.com/PaddlePaddle/Paddle/pull/2115
  • PR: https://github.com/PaddlePaddle/Paddle/pull/2083

• PaddlePaddle cloud:

  • PR:https://github.com/PaddlePaddle/cloud/pull/30

Gang Liao

• Add CMake to port Majel's DArray
• Add Majel in Paddle Design Doc https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/majel
• Implement cc_xxx and nv_xxx to simplify CMake for C++ and CUDA https://github.com/PaddlePaddle/Paddle/pull/2174
• Implement go_xxx to simplify CMake for Golang https://github.com/PaddlePaddle/Paddle/pull/2182
• Add some unit test to verify above features

typhoonzero(wuyi)

• Paddle cloud:
  • A command-line client for submitting jobs to paddle cloud. Support v1 API currently, and it’s able to port to v2 simply.
  • Internal CPU kubernetes cluster, mount HDFS to nodes and copy HDFS data to do a large scale training job test. When this is done, we can try to migrate some of the MPI training jobs on it. Depends on above command-line client.
  • TODO: Enable RBAC and quota. Embed file upload.
• MPI Cluster bug when training with >150 nodes
  • Found a solution, testing.
• [WIP]Paddle v2 API distributed training
  • https://github.com/PaddlePaddle/Paddle/pull/2072 Thanks for @luotao ’s comment. Will update the doc later.
• PR reviews:
  • https://github.com/PaddlePaddle/Paddle/pull/2172
  • https://github.com/PaddlePaddle/cloud/pull/26
  • https://github.com/PaddlePaddle/cloud/pull/30

hedaoyuan

• Embedded
  Face model deployment progress: The Face model has been transformed into a Paddle model and predicted by CAPI.But the computational performance of the model inference can not meet the requirements. See agroup documentation for detailed data.
  http://agroup.baidu.com/paddlepaddle/md/article/319536?side=folder

Reconstructing convolutional modules to improve model inference performance.
https://github.com/PaddlePaddle/Paddle/issues/2177
https://github.com/hedaoyuan/Function

• Bug & Review
  https://github.com/PaddlePaddle/Paddle/issues/2101
  https://github.com/PaddlePaddle/Paddle/pull/1665#discussion_r116676770

Yancey1989(yanxu)

• Paddle cloud:
  • Provide a RESTful API for submitting Paddle Job on Paddle cloud,PR is here
  • [WIP]Transfer a MPI traininig job to Kubernetes cluster.
• Code Reivews:
  • Design doc about FileManager
  • Design doc about Paddle cloud client

luotao

• fix some issus:
  • fix travis-ci of docs: #2175
  • add assert when is_seq=True, there must be a bootLayer in the network: #2161
  • add appointments for contributing codes in document: #2139
  • add multiple outputs in annotation of paddle.infer: #2136
  • add explanation of why paddle version 0.0.0 in faq: #2116
  • fix error cmake infomation: #2114
  • modify install url in all books: #315
• code review on docs and models:
  • https://github.com/PaddlePaddle/Paddle/pull/2072#pullrequestreview-38282810
  • https://github.com/PaddlePaddle/models/pull/24#pullrequestreview-38288110
  • https://github.com/PaddlePaddle/models/pull/36#pullrequestreview-38575791
• Wechat PaddlePaddle: 126 fans -> 150 fans

LiuYiqun

• Add dockfile for cross-compiling for Android: https://github.com/PaddlePaddle/Paddle/pull/2180
• Add documentations for cross-compiling: https://github.com/PaddlePaddle/Paddle/pull/2179
• Try to use hardfp for Android, and compare the performance's difference

chengxiaohua1105

• Paddle demo:
  • finish updatinging ngrm code for reviewing

xushaoyong

• HomePage：
  • Release the new homepage http://paddlepaddle.org/
  • add a link to older version docs on paddle's homepage https://github.com/PaddlePaddle/Paddle/issues/2159
• Regression model
  • add new dataset, modify the readme docs

fengjiayi

• Paddle refactoization:
  • read Majel code
  • try importing Struct Dim of Majel into Paddle
  • PR: https://github.com/PaddlePaddle/Paddle/pull/2144
• Model:
  • PR merged: https://github.com/PaddlePaddle/models/pull/24

qijun

• New tensor library in Paddle:
  • add detailed docs about porting Majel to Paddle
  • PR: https://github.com/PaddlePaddle/Paddle/pull/2176
  • try importing Allocation of Majel into Paddle

Xinghai Sun

• Models:
  • PR: update "MT with NTM" model, complete docs (https://github.com/PaddlePaddle/models/pull/36/commits).
• Deep Speech 2:
  • Issue: investigation and task breakdown (https://github.com/PaddlePaddle/models/issues/44)
• Bug reports:
  • https://github.com/PaddlePaddle/Paddle/issues/2141
  • https://github.com/PaddlePaddle/Paddle/issues/2142
  • https://github.com/PaddlePaddle/Paddle/issues/2186

guosheng

• discussions and document writing for chunk evaluator and testing the fixed evaluator by @pkuyym
• update configuration and README of NER model
• start to work on the tool for transforming models from Caffe to Paddle

yangyaming

• Bugs:
  • Enrich ChunkEvaluator to expose precision, recall and F1-score to user
  • Revise usage document for v2 api of ChunkEvaluator @guosheng
  • PR: https://github.com/PaddlePaddle/Paddle/pull/2165
• Features:
  • Modify Matrix to support circular convolution for sequence data
  • Enrich data type for unitest to enable testing for ConvShift Layer
  • PR: https://github.com/PaddlePaddle/Paddle/pull/2133

dongzhihong

develop the new optimizer lib for training and ParameterServer https://github.com/PaddlePaddle/Paddle/pull/2190

parameterServer related discussion https://github.com/PaddlePaddle/Paddle/issues/2169 https://github.com/PaddlePaddle/Paddle/issues/2168 https://github.com/PaddlePaddle/Paddle/issues/2106

Yu Yang

Design Doc:

• How to refine v2 API implementation to fix current defects.
  • https://github.com/PaddlePaddle/Paddle/pull/2104
  • demo implemetation https://github.com/PaddlePaddle/Paddle/pull/2096
• How to refine Layer/Projection registration ways, to refactor the topology in CPP
  • https://github.com/PaddlePaddle/Paddle/pull/1665
  • implemetation https://github.com/PaddlePaddle/Paddle/pull/2087

NHZLX(Zhaolong Xing)

• issue:
  • https://github.com/PaddlePaddle/Paddle/issues/2183
• Embedded:
  • a survey of ICLR 17's paper on model compression about pruning, paper is here.

Dang Qingqing

1. Image
   • Read paper about image segmentation and plan to support this feature.
2. Read some paper about deep-speech
3. PR Review
   • https://github.com/PaddlePaddle/models/pull/31
   • https://github.com/PaddlePaddle/Paddle/pull/2178

Yibing Liu

• Models:
  • Complete README of NMT: https://github.com/PaddlePaddle/models/pull/22
• Paddle:
  • Resolving the problem that cannot set gradient clipping and some other hyper parameters globally.

juliecbd

• Pull Request:
  • https://github.com/PaddlePaddle/book/pull/316
• Build paddle in Mac
• Found issues in python examples in paddle/demo : seems like compatibility issue. need review