RubyKaigi 2018 LT に出していた CFP が accept されましたので、 Red Chainer と Numo::NArray のパフォーマンス改善の取り組み (下記の二つ)について、 どんな事をしたか、どれくらい改善されたのかを python Chainer と numpy の結果と比較しながら、お話させて頂く予定です。
- https://github.com/red-data-tools/red-chainer/pull/48
- https://github.com/ruby-numo/numo-narray/pull/94
LT は初日 (5/31) になります。
Red Chainer の紹介は @hatappi さんが同じ日にメインセッションの Deep Learning Programming on Ruby - RubyKaigi 2018 でお話しされますので、そちらをご覧ください。
受理された CFP はこんな感じです。ご参考までに。
## Improve Red Chainer and Numo::NArray performance ## Abstract I would like to talk my progress of daily contribution to Ruby world. I'm planing to talk the performance improvement to Red Chainer and Numo::NArray(used by Red Chainer) gem. I hope that machine learning will be comfortable to use with Ruby. ## Details I would like to talk my progress of daily contribution to Ruby world. Last year, I talked about "How to get on the shoulder of the giant for machine learning" at RejectKaigi 2017. I'm planing to talk the performance improvement to Red Chainer and Numo::NArray(used by Red Chainer) gem as following with 5 minutes at RubyKaigi 2018 LT. ## Red Chainer - https://github.com/red-data-tools/red-chainer/pull/48 When my pull request will be merged, red-chainer gem performance will improve as follows. * examples/mnist.rb 1 epoc -> 73 % faster ## Numo::NArray - https://github.com/ruby-numo/numo-narray/pull/94 With my pull request, the performance of Numo::NArray gem has been improved as follows. * x.inplace + y or - y or * y -> 10-15 % faster * x.inplace + 1 or - 1 or * 1 -> 2.1 times faster * x.inplace / y or / 1 -> 3.2-3.8 times faster Thanks.
