2017 10 04

wangkuiyi

• Python API Design Doc imporvement: https://github.com/reyoung/Paddle/pull/6
• Paddle Cloud UI design: https://github.com/PaddlePaddle/cloud/pull/378#pullrequestreview-66318158
• TensorArray design: https://github.com/PaddlePaddle/Paddle/pull/4564#pullrequestreview-66645688
• farm
  • Continuous integration https://github.com/PaddlePaddle/farm/issues/3
  • Unit tests
    • https://github.com/PaddlePaddle/farm/issues/2
    • https://github.com/PaddlePaddle/farm/pull/4#pullrequestreview-66841276
  • Branching and releasing model https://github.com/PaddlePaddle/farm/issues/1
• Update LoDTensor design doc https://github.com/PaddlePaddle/Paddle/pull/4579
• Executor https://github.com/PaddlePaddle/Paddle/pull/4537#pullrequestreview-66923299
• Why adding type_defs.h https://github.com/PaddlePaddle/Paddle/issues/4587
• GetCUDADeviceCount https://github.com/PaddlePaddle/Paddle/pull/4595

reyoung/Yu Yang

• Generate gradient operators when compile-time.

  • Design doc https://github.com/PaddlePaddle/Paddle/pull/4517
  • Step 1. Basic Classes https://github.com/PaddlePaddle/Paddle/pull/4551
  • Step 2. [Done, UnderReview] https://github.com/PaddlePaddle/Paddle/pull/4566

• Change PADDLE_ONLY_CPU -> PADDLE_WITH_GPU

  • https://github.com/PaddlePaddle/Paddle/pull/4584

• Simplize sum_op, compose its gradient operator

  • https://github.com/PaddlePaddle/Paddle/pull/4562
  • Since there are many kernels started in sum_grad, it is not fast either.

• Unify map for OpDescBind and OperatorBase

  • https://github.com/PaddlePaddle/Paddle/pull/4547

• Remove add_op since it is duplicated with sum_op

  • https://github.com/PaddlePaddle/Paddle/pull/4514

• Use double precision to stablize some gradient checks

  • https://github.com/PaddlePaddle/Paddle/pull/4491

• Simplize Activation in Paddle

  • https://github.com/PaddlePaddle/Paddle/pull/4485

• Stablize gradient test of elementwise-mul

  • https://github.com/PaddlePaddle/Paddle/pull/4478

• Remove OperatorBase::InferShape

  • https://github.com/PaddlePaddle/Paddle/pull/4458

• Fix MacOS Compilation error

  • https://github.com/PaddlePaddle/Paddle/pull/4560

dongzhihong

• Refactorization:

  • Implement the save/load/checkpoint support
    • https://github.com/PaddlePaddle/Paddle/pull/4602
  • fix add operator bug in backward process
    • https://github.com/PaddlePaddle/Paddle/pull/4548
  • remove default delete copy construct method
    • https://github.com/PaddlePaddle/Paddle/pull/4550
  • Discuss Executor design with @qijun, @helinwang -https://github.com/PaddlePaddle/Paddle/pull/4537

• Reviews

  • Sigmoid Cross Entropy
    • https://github.com/PaddlePaddle/Paddle/pull/4448
  • Leaky Relu activation
    • https://github.com/PaddlePaddle/Paddle/pull/4604
  • Conv Shift operator
    • https://github.com/PaddlePaddle/Paddle/pull/4591
  • Rmsprop operator
    • https://github.com/PaddlePaddle/Paddle/pull/4565
  • Changing SGD operator
    • https://github.com/PaddlePaddle/Paddle/pull/4586
  • fix bug in backward process
    • https://github.com/PaddlePaddle/Paddle/pull/4582
  • guide AML team to send PRs to improve our documents
    • https://github.com/PaddlePaddle/Paddle/pull/4484
    • https://github.com/PaddlePaddle/Paddle/pull/4549
    • https://github.com/PaddlePaddle/Paddle/pull/4479

qiaolongfei

refactoring

• remove Runtime InferShape for cond op
  • https://github.com/PaddlePaddle/Paddle/pull/4518
• add some check to operator.run
  • https://github.com/PaddlePaddle/Paddle/pull/4544
• use float32 in cond_op
  • https://github.com/PaddlePaddle/Paddle/pull/4546
• refactor rnn infershape
  • https://github.com/PaddlePaddle/Paddle/pull/4553

code review

• dynamic recurrent op forward c++ implentation
  • https://github.com/PaddlePaddle/Paddle/pull/4597
• dynamic recurrent op forward c++ implentation
  • https://github.com/PaddlePaddle/Paddle/pull/4597
• Implementing the Adamax optimizer operator
  • https://github.com/PaddlePaddle/Paddle/pull/4538
• Implementing the Adagrad optimizer step operator
  • https://github.com/PaddlePaddle/Paddle/pull/4558
• Changing learning rate from attribute to input(float)
  • https://github.com/PaddlePaddle/Paddle/pull/4568
• Adding Adadelta optimization operator
  • https://github.com/PaddlePaddle/Paddle/pull/4576
• Changing learning rate from type Input(float) to Input(tensor)
  • https://github.com/PaddlePaddle/Paddle/pull/4578