Projects

Open Source Contributions

Projects

Open Source Contributions

• online-judge-tools/oj (Python)

  • Skip language option tags without value at AtCoder

• treasure-data/digdag (Java)

  • #708 Add dependencies for copied tasks for grouping task retry
  • #845 Implement persistent params operator
  • #850 [Experimental] Show logs by task
  • #883 Implement td_result_export> operator
  • #880 Make page_size configurable

• google/google-auth-library-ruby (Ruby)

  • #97 Retry http request when unexpected error occurs
  • #98 Fix rubocop warning

• openid/ruby-openid (Ruby)

  • #103 Update admin/mkassoc to enable to work

• phstc/shoryuken (Ruby)

  • #177 Allow symbol as a queue name in shoryuken_options

• blue-oil/blueoil (C++)

  • #427 [refactor] remove a redundant semicolon from dlk src/func/conv2d.cpp

Projects

chainer_prefetch_multiprocess_iterator (Python)

• This is the reference implementation of my study “Accelerating Machine Learning I/O by Overlapping Data Staging and Mini-batch Generations”.
• This is a Chainer Iterator class that executes prefetching training data from slow storages (such like parallel file systems) into fast storage (such as SSD) and generating mini-batches in the same time.
• The aim of this study is to conceal the time for staging-in training dataset into node-local storages in computation nodes at HPC clusters (such as ABCI, TSUBAME, Cygnus, and so on).

chainer_minibatch_size_optimizer (Python)

• This is the reference implementation of my paper, 深層ニューラルネットワークにおける訓練高速化のための自動最適化.
• This is a Chainer Trainer extension class that optimizes mini-batch size adaptically for training a model with one GPU.

inference_engine (C++)

• This is an ONNX runtime implementation, such like onnxruntime or menoh.
• For now, Gemm, Conv, MaxPool, Relu, Softmax, Dropout, Reshape are supported.
• All backend implementation is my own implementation for CPU (This means that current backend implementation does not use optimized matrix libraries, such like Blas, Intel MKL-DDN, and so on.).

simple_map_reduce - Distributed MapReduce framework (Ruby)

• This is an experimental project to study MapReduce.
• The blog post to illustrate this product (Japanese).

optimization_experiments (C)

• This is an experimental project to study optimization for GEMM and GEMV.
• This projects includes dgemm and dgemv implementations which are optimized by loop exchange, loop unloop, blcoknize, padding repeatedly.

convolution_experiments (Python)

• This is an experimental project to study implementation for Convolution with direct and im2col style repeatedly.

config2args (Rust)

• This is a CLI tool to convert json config file into GNU CLI option style (such like --key1 value1 --key2 value2)