Ruby のデータサイエンス対応の現状確認

Meguro.rb 2017.11.30

Transcript

1. Ruby ךر٦ة؟؎ؒٝأ㼎䘔ך
   植朐然钠
   Kenta Murata, Speee Inc. Ꟛ涪鿇 R&D ؚٕ٦ف
   Thu, Nov 30, 2017 Meguro.rb
   1 / 16
   

   View Slide

2. self.introduce
   Kenta Murata (@mrkn)
   Full-time CRuby committer at Speee Inc.
   CRuby
   The maintainer of bigdecimal
   Other OSS
   pycall.rb, mxnet.rb, enumerable-statistics, vim-cruby
   Hobby
   Play with my daughter, Mathematics, Physics, Computer Science,
   Reading SF, etc.
   2 / 16
   

   View Slide

3. 植㖈ծRuby ךر٦ة؟؎ؒٝأ㼎䘔כ麓床劍דׅկ
   劤傈כծו׿זفٗآؙؑزָ㶷㖈׃ծ׉׸׊׸ו׿ז暴䗙ָ
   ֮׷ַ׾稱➜׃תׅկ
   ׫זׁ׿כծ׈מוַֿךفٗآؙؑزח⿫⸇׃גְ׋׌ֹծ
   Ruby ךر٦ة؟؎ؒٝأ㼎䘔׾⸇鸞ׅ׷䔲✲罏חז׏גְ׋׌
   ֹ׋ְדׅկ
   3 / 16
   

   View Slide

4. Contents
   SciRuby
   Ruby Numo
   pycall.rb
   mxnet.rb
   Red Data Tools project
   GPGPU ΁ͷରԠঢ়گ
   RubyData organization
   4 / 16
   

   View Slide

5. SciRuby
   NMatrix Λத৺ͱ͢ΔՊֶٕज़ܭࢉ༻ gem ܈
   ւ֎Ͱ࠷΋༗໊ͳ Ruby ͷՊֶٕज़ܭࢉϥΠϒϥϦ܈
   ൃىਓͨͪ͸࠷ۙ͸શ͘։ൃ͓ͯ͠ΒͣɺGSOC ͷϓϩδΣΫτͰֶੜ͕
   ৭ʑ࡞͍ͬͯΔ΋ͷͷدͤूΊঢ়ଶʹͳ͍ͬͯΔ
   Ұ෦ͷجຊԋࢉ͕஗͍όά͕͋Δ
   5 / 16
   

   View Slide

6. Ruby Numo
   Numo::NArray Λத৺ͱ͢ΔՊֶٕज़ܭࢉ༻ gem ܈
   ݩ૆ NArray ͷ࡞ऀͰ͋Δాதণ޺͞Μ͕ൃىਓͰ͋Γ։ൃͷத৺ʹ͍Δ
   RubyKaigi Ͱ΋ൃද͞Ε͍ͯͯɺ೔ຊͰ͸ͦͦ͜͜஌ΒΕ͍ͯΔͱࢥ͏
   ւ֎Ͱ͸͋·Γ஌ΒΕ͍ͯͳ͍ (?)
   6 / 16
   

   View Slide

7. SciRuby ה Ruby Numo דדֹ׷ֿה
   σʔλϑϨʔϜϥΠϒϥϦ Daru ͕ NMatrix ʹରԠ͍ͯ͠Δ
   [HELP] ୭͔ Numo::NArray ʹ΋ରԠ͍ͤͯͩ͘͞͞
   Պֶٕज़ܭࢉ༻ͷجૅؔ਺܈
   GSL ͷόΠϯσΟϯά͕ SciRuby ͱ Ruby Numo ͷͦΕͧΕʹରͯ͠
   ଘࡏ͢Δ
   ՄࢹԽ
   Gnuplot Λ࢖͏ϥΠϒϥϦ͕ SciRuby ͱ Ruby Numo ͷͦΕͧΕʹର
   ͯ͠ଘࡏ͢Δ
   rbplotly ͕྆ํʹରԠ͍ͯ͠Δ (͸ͣ)
   7 / 16
   

   View Slide

8. pycall.rb
   Ruby ͔Β Python Λݺͼग़ͤΔ
   Ruby ͱಉ͡ϓϩηε಺Ͱ Python ΠϯλʔϓϦλΛॳظԽ͠ݺͼग़͢
   ࣍ͷϦϦʔε͔Β Python Λݺͼग़͢ࡍʹ GVL Λղ์͢ΔΑ͏ʹͳΔͷ
   Ͱɺpuma Ͱಈ͘ Rails ΞϓϦέʔγϣϯͰ΋࢖͑Δʂ
   ͔͠͠ɺPython ଆͷ GIL ͷ੍ޚΛ·ͩ΍ͬͯͳ͍ͷͰɺRuby ଆͷෳ਺ͷ
   εϨου͔Β Python ͷػೳΛݺͼग़ͯ͠͸ͳΒͳ͍
   ཉ͍͠ػೳͷఏҊΛϓϧϦΫΤετͷܗͰืू͍ͯ͠·͢
   Ұॹʹ։ൃͯ͘͠ΕΔɺRuby ͱ Python ͷ྆ํʹৄ͍͠ਓ΋ืूͯ͠·͢
   8 / 16
   

   View Slide

9. mxnet.rb
   MXNet ͷ Ruby όΠϯσΟϯά
   10݄ࠒ͔Βຊؾग़ͯ͠ RubyConf ·Ͱʹ׬੒ͤ͞Α͏ͱ͚ͨ͠ͲແཧͰ͠
   ͨʂʂ
   3݄ͷԭೄ RubyKaigi 02 ·ͰʹͳΜͱ͔͢Δ
   Ruby ͱ MXNet ͷ྆ํʹৄ͍͠ਓͷڠྗΛืू͍ͯ͠·͢
   9 / 16
   

   View Slide

10. Red Data Tools project
    Apache Arrow Λத৺ͱͨ͠σʔλπʔϧΤίγεςϜͷ Ruby ൛
    ਢ౻͞Μ্ཱ͕ͪ͛ͨ
    ਢ౻͞Μ͸ Apache Arrow ͷ PMC ʹ΋ͳͬͨ
    Red Chainer ͳͲɺApache Arrow ͱ͸௚઀ؔ܎ͳ͍೿ੜϓϩμΫτͷ։
    ൃ΋࢝·͍ͬͯΔ
    ͓ͦΒؔ͘܎ऀ͸શһ͕೔ຊਓͰ͋Δ
    ຖ݄ʮOSS Gate ౦ژϛʔτΞοϓ for Red Data ToolsʯͰ։ൃ͕ਐΊΒΕ
    ͍ͯΔ
    12݄ͷΠϕϯτͷҊ಺ΛޙͰ͠·͢
    10 / 16
    

    View Slide

11. GPGPU פך㼎䘔朐屣
    tensorflow.rb Ͱ GPU ͕࢖͑Δ (͸ͣ)
    pycall.rb Ͱ Python ͷ GPGPU ରԠϥΠϒϥϦΛ࢖͑Δ
    ࠓ೥ 2017 ೥ͷ GSOC Ͱ ArrayFire ͷόΠϯσΟϯά͕࡞ΒΕͨ
    ࠓ೥ 2017 ೥ͷ Ruby Grant Ͱ GPGPU ରԠϓϩδΣΫτ͕2݅࠾୒
    1. RbCUDA: ArrayFire ͷόΠϯσΟϯάΛ࡞ͬͨਓͷϓϩδΣΫτͰɺ
    CUDA ϥϯλΠϜϥΠϒϥϦͷ Ruby όΠϯσΟϯά
    2. cumo: sonots ͞ΜͷϓϩδΣΫτͰɺcupy ͷ Ruby ൛Λ
    Numo::NArray ʹରͯ͠࡞Δ
    mxnet.rb ͕࢖͑ΔΑ͏ʹͳΔͱɺ͜Ε΋ GPGPU ʹରԠ͢Δ
    11 / 16
    

    View Slide

12. Ruby דر٦ة؟؎ؒٝأׅ׷橆㞮׾知⽃
    ח鑐ׅ倯岀
    ࣮͸ Docker image Λ࡞ͬͯ͋Δɻ
    docker run --rm -it -p 8888:8888 rubydata/notebooks
    ͜ΕͰɺJupyter notebbok ্ཱ͕͕ͪΔͷͰɺIRuby kernel ͷϊʔτϒοΫ
    Λ࡞Ε͹ࢼͤΔɻ
    12 / 16
    

    View Slide

13. RubyData organization
    SciRubyɺRuby NumoɺRed Data ToolsɺͳͲෳ਺ͷϓϩδΣΫτʹਓ͕
    ෼ࢄ͍ͯ͠Δݱঢ়͸ɺ७ਮͳར༻ऀ͕ށ࿭͍΍͍͢
    ϓϩδΣΫτʹؔ܎ͳ͘ɺRuby ͱσʔλπʔϧΛ૊Έ߹Θͤͯ࢖͓͏ͱ͠
    ͍ͯΔਓ͕ू·ͬͯ৘ใަ׵͕Ͱ͖Δ৔͕ඞཁͩΖ͏
    ͱ͍͏Θ͚Ͱɺࢲ͕࡞Γ࢝Ί·ͨ͠
    http://ruby-data.org/
    ·ͩϥϯσΟϯάϖʔδ͔͠ͳ͍ʂʂ
    13 / 16
    

    View Slide

14. RubyData needs helps!!
    ҎԼͷΑ͏ͳਓΛืूͯ͠·͢
    ίϛϡχςΟӡӦ (ਓͷू·ΓΛ੍ޚ͢Δ͜ͱ) ͕ಘҙͳਓ (ࢲ mrkn ͕
    Ұ൪ෆಘҙͳ͜ͱ)
    AWS ʹৄ͍͠ਓ (discourse ͷ্ཱͪ͛ͱӡӦΛख఻ͬͯཉ͍͠)
    ࣗવݴޠͰͷܧଓతͳ৘ใൃ৴͕ಘҙͳਓ (ʮRuby ͰσʔλαΠΤϯ
    εʯతͳ৘ใൃ৴Λεέʔϧ͍ͤͨ͞)
    ӳޠ͕ಘҙͳਓ (๻ͨͪͷมͳӳޠΛ௚ͯ͠ཉ͍͠)
    docker-stacks ͷҡ࣋ͱൃలΛੵۃతʹ΍ͬͯ͘ΕΔਓ
    14 / 16
    

    View Slide

15. ׫זׁ׿ָ➙ּׅדֹ׷ֿה
    1. ࢖ͬͯόάใࠂ
    2. ೔ຊʹดͯ͡ΔϓϩμΫτΛւ֎޲͚ʹӳޠͰ঺հ͢Δ
    3. ར༻ऀ͕૿͑ΔΑ͏ͳϒϩάهࣄΛॻ͘ (΍ͬͯΈͨܥ)
    4. ։ൃऀ͕૿͑ΔΑ͏ͳϒϩάهࣄΛॻ͘ (։ൃ؀ڥΛ࡞ΔɺίʔυΛಡΜͰ
    Έͨܥ)
    5. طଘͷϥΠϒϥϦʹϓϧϦΫΛૹͬͯ։ൃʹࢀՃ͢Δ
    6. RubyData ͷӡӦνʔϜʹೖͬͯಘҙͳ͜ͱΛ΍Δ
    7. OSS Gate for Red Data Tools ʹࢀՃͯ͠։ൃ͢Δ
    15 / 16
    

    View Slide

16. OSS Gate for Red Data Tools in Speee 2017.12.19
    Red Data Tools ϓϩδΣΫτͷ։ൃऀ͕ू·Γ·͢ɻ
    ৔ॴ: ࿡ຊ໦4-1-4 ࠇ࡚Ϗϧ 5F Speee Inc. ηϛφʔϧʔϜ
    ࣌ؒ: 2017೥12݄19೔
    19:00 ։৔
    19:30 ։࢝
    https://speee.connpass.com/event/72926/
    16 / 16
    

    View Slide