Rubyで機械学習が出来る 未来を目指すRed Data Toolsの 現状と今後について
by
hatappi
×
Copy
Open
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Slide 1
Slide 1 text
RubyͰػցֶश͕ग़དྷΔ ະདྷΛࢦ͢Red Data Toolsͷ ݱঢ়ͱࠓޙʹ͍ͭͯ by Yusaku Hatanaka @RubyWorld Conference 2017 1
Slide 2
Slide 2 text
self.intoroduction { name: "Yusaku Hatanaka", twitter: "@hatappi", github: "hatappi", hatena: "hatappi1225", company: "Speee, Inc." languages: %w(ruby go python), icon: "ɹ " } 2
Slide 3
Slide 3 text
ࣗݾհ w ാத༔࡞ w 5XJUUFSɺ(JUIVCIBUBQQJ ͯͳϒϩάIBUBQQJ w גࣜձࣾ4QFFF w σδλϧίϯαϧςΟϯάࣄۀຊ෦ ΞυςΫࣄۀ෦6;06ࣄۀ w 6;06ࣄۀΤϯδχΞ
Slide 4
Slide 4 text
Agenda • Rubyʹ͓͚Δػցֶशͷݱঢ়ʹ͍ͭͯ • Red Data Toolsʹ͍ͭͯ • ݱঢ়ͷऔΓΈ • ະདྷʹ͍ͭͯ • ·ͱΊ 4
Slide 5
Slide 5 text
Agenda • Rubyʹ͓͚Δػցֶशͷݱঢ়ʹ͍ͭͯ • Red Data Toolsʹ͍ͭͯ • ݱঢ়ͷऔΓΈ • ະདྷʹ͍ͭͯ • ·ͱΊ 5
Slide 6
Slide 6 text
6 RubyͰ ػցֶश͍ͨ͠ʂ
Slide 7
Slide 7 text
7 ػցֶशʹݶΒͣ σʔλੳʹ͓͍ͯ σʔλ࿈ܞඞཁͱͳΔ
Slide 8
Slide 8 text
8 σʔλ࿈ܞͱ͍͏ͱ JSONCSV ??
Slide 9
Slide 9 text
JSONCSV ϑΝΠϧͰͷΓͱΓ͚ͩͰͳ͘APIͰͷϦΫΤε τɾϨεϙϯεͳͲͰ༻͞ΕΔ 9 PythonͰग़ྗͨ͠CSVΛRubyͰಡΈࠐΉྫ write: 0.8009s read: 0.1227s
Slide 10
Slide 10 text
CSVJSONͰ େ͖ͳσʔλΛॲཧ ಛఆͷྻͰूܭॲཧΛ͍ͨ͠߹Ͱ Ұͯ͢ΛಡΈࠐΉඞཁ͕͋ΔͷͰ େ͖ͳσʔλΛॲཧ͢Δʹݶք͕͋Δ 10 ػցֶशͰඦສͷେ͖ͳσʔληοτΛ͏͜ͱ ͋ΔͷͰେ͖ͳσʔλΛѻ͑ΔΑ͏ʹͳΓ͍ͨ
Slide 11
Slide 11 text
11 Apache Parquet
Slide 12
Slide 12 text
Apache Parquetͱ • ΧϥϜܕͷσʔλϑΥʔϚοτ • શ෦ΛಡΈࠐ·ͣͱσʔλͷҰ෦ΛऔΓग़ͤΔ • ྫ͑ಛఆͷྻΛऔΓग़ͯ͠ूܭ͢Δ • αΠζ͕খ͍͞ • ΧϥϜͰσʔλΛ֨ೲ͢ΔͨΊಉ͡ܕͷσʔλΛѹॖ͢ Δ͜ͱ͕ग़དྷΔ 12 େ͖ͳσʔλΛॲཧͯ͠อଘ͢Δ͜ͱʹ͍͍ͯΔ
Slide 13
Slide 13 text
Apache Parquet 13 PythonͰग़ྗͨ͠ParquetΛRubyͰಡΈࠐΉྫ write: 0.1390s read: 0.0422s
Slide 14
Slide 14 text
14 େྔͷσʔλΛ ѻ͑ΔΑ͏ʹͳͬͨ
Slide 15
Slide 15 text
ෳͷγεςϜؒͰͷ σʔλΛ࿈ܞ ParquetσʔλΛѹॖ͢ΔͨΊσʔλॲཧ͢Δաఔ ͰෳγεςϜؒͰσʔλ࿈ܞΛߦ͏ࡍͦΕΛݩʹ ͢Φʔόʔϔου͕ੜ͡Δ 15 γʔϜϨεʹγεςϜؒͰσʔλ࿈ܞ͍ͨ͠!
Slide 16
Slide 16 text
16 Apache Arrow
Slide 17
Slide 17 text
Apache Arrowͱ • ΠϯϝϞϦͰΧϥϜܕσʔλΛѻ͏ͨΊͷϑΥʔϚοτͱΞϧΰ ϦζϜ • γϦΞϥΠζɾσγϦΞϥΠζίετ͕΄΅0 • σʔλΛ͢ଆड͚औΔଆCPUΛΘͳ͍ • θϩίϐʔ • σʔλΛड͚औΔଆϝϞϦίϐʔʹ࣌ؒΛΘͳͯ͘Α͍ 17 σʔλަʹΉ͍͍ͯΔ
Slide 18
Slide 18 text
Apache Arrow 18 PythonͰग़ྗͨ͠ParquetΛRubyͰಡΈࠐΉྫ write: 0.0775s read: 0.0074s
Slide 19
Slide 19 text
19 1ZUIPOͰͷॻ͖ࠐΈ T 3VCZͰͷಡΈࠐΈ T $47 "QBDIF1BSRVFU "QBDIF"SSPX 5000Ϩίʔυʹର͢Δ ಡΈॻ͖ൺֱ·ͱΊ
Slide 20
Slide 20 text
Apache Arrow͕ͳ͍࣌ 20
Slide 21
Slide 21 text
Apache Arrow͕͋Δ࣌ 21
Slide 22
Slide 22 text
RubyͰParquetΛѻ͑Δͱ 22 ෳͷγεςϜؒ"QBDIF "SSPXͰσʔλΛ࿈ܞΛߦ͍ σʔλੳΛߦ͏ ParquetΛѻ͏͜ͱͰσʔλͷ࿈ܞग़དྷͨ ͔͠͠RubyͰσʔλੳ͕͍ͨ͠ʂʂʹಧ͔ͳ͍
Slide 23
Slide 23 text
RubyͰArrowΛѻ͑Δͱ 23 ෳͷγεςϜؒ"QBDIF "SSPXͰσʔλΛ࿈ܞΛߦ͍ σʔλੳΛߦ͏ RubyͰσʔλॲཧͷҰ෦Λ୲͏͜ͱ͕ग़དྷΔ!!
Slide 24
Slide 24 text
3VCZͰ"SSPXΛѻ͑Δͱ 24 "QBDIF"SSPXʹରԠ͢Δ͜ͱͰඞཁͳ෦͔Β 3VCZΛͬͨσʔλੳΛ͡ΊΔ͜ͱ͕ग़དྷΔ w 3VCZͰूΊͨσʔλΛ"QBDIF"SSPXʹରԠͯ͠ ͍Δ1BOEBT4QBSLʹ࿈ܞ͠ੳͨ݁͠ՌΛ3VCZ Ͱड͚औͬͯ3BJMTΛͬͨXFCΞϓϦͰՄࢹԽ w ޙʹ1BOEBT4QBSL෦Λঃʑʹ3VCZҠߦ͢ Δ͜ͱग़དྷΔ ྫ
Slide 25
Slide 25 text
25
Slide 26
Slide 26 text
Agenda • Rubyʹ͓͚Δػցֶशͷݱঢ়ʹ͍ͭͯ • Red Data Toolsʹ͍ͭͯ • ݱঢ়ͷऔΓΈ • ະདྷʹ͍ͭͯ • ·ͱΊ 26
Slide 27
Slide 27 text
Red Data Tools • גࣜձࣾΫϦΞίʔυͷਢ౻͞Μ͕20172݄ʹ ϓϩδΣΫτΛઃཱ • Ruby༻ͷσʔλॲཧπʔϧΛఏڙ͢Δ͜ͱΛత ͱͨ͠ϓϩδΣΫτ • ଟ͘ͷݴޠ͕ڞ௨ͯ͠༻Ͱ͖ΔApache ArrowΛ ༻͢Δ͜ͱͰRubyίϛϡχςΟʔΛ͑ͯڠྗ ͢Δ 27
Slide 28
Slide 28 text
Agenda • Rubyʹ͓͚Δػցֶशͷݱঢ়ʹ͍ͭͯ • Red Data Toolsʹ͍ͭͯ • ݱঢ়ͷऔΓΈ • ະདྷʹ͍ͭͯ • ·ͱΊ 28
Slide 29
Slide 29 text
ݱঢ়ͷऔΓΈ • Red Arrow • طଘgemͷArrowରԠ • ৽͍͠πʔϧΛఏڙ 29
Slide 30
Slide 30 text
ݱঢ়ͷऔΓΈ • Red Arrow • طଘgemͷArrowରԠ • ৽͍͠πʔϧΛఏڙ 30
Slide 31
Slide 31 text
Red Arrow w "QBDIF"SSPXͷ3VCZόΠϯσΟϯά w (JU)VCSFEEBUBUPPMTSFEBSSPX w "SSPX(-JCͱHPCKFDUJOUSPTQFDUJPOΛͬͯ "QBDIF"SSPXͷόΠϯσΟϯάΛ࣮ݱ͍ͯ͠Δ 31
Slide 32
Slide 32 text
ݱঢ়ͷऔΓΈ • Red Arrow • طଘgemͷArrowରԠ • ৽͍͠πʔϧΛఏڙ 32
Slide 33
Slide 33 text
SciRubyɾRuby Numo • Պֶٕज़ܭࢉɺσʔλՄࢹԽ༻్ͷGem܈Λ։ൃ ͍ͯ͠ΔϓϩδΣΫτ • σʔλϑϨʔϜΛѻ͑Δdaruߦྻܭࢉ͕ग़དྷΔ numo-narryͳͲ͕͋Δ • ֤༻్Ͱݸผͷgemଘࡏ͢Δ͔Β ͦΕΒ͕࿈ܞͰ͖ΔΑ͏ʹͳΔͱ ͬͱΑ͍ʂ 33
Slide 34
Slide 34 text
PyCall • @mrkn͞Μ͕࡞͞Ε͍ͯΔRubyͱPythonͷϒ ϦοδϥΠϒϥϦ • PythonͰ࡞͞Εͨطଘͷࢿ࢈Λͬͯ PythonͷΦϒδΣΫτΛ RubyͰ༻͢Δ͜ͱ͕ग़དྷΔ 34
Slide 35
Slide 35 text
طଘgemͷArrowରԠ 35 4DJ3VCZ 1Z$BMM "QBDIF"SSPXʹରԠͤ͞Δ͜ͱͰ طଘͷHFNΛ༻ͯ͠σʔλੳΛ͡ΊΒΕΔ 3VCZ/VNP
Slide 36
Slide 36 text
ݱঢ়ͷऔΓΈ • Red Arrow • طଘgemͷArrowରԠ • ৽͍͠πʔϧΛఏڙ 36
Slide 37
Slide 37 text
Red Chainer • ChainerΛRubyϙʔςΟϯάͨ͠ͷ • ChainerͷΫϥεύϥϝʔλͷ࣋ͪํΛࢀߟʹ RubyͰॻ͘͜ͱͰ0͔Β࡞ΔͷͰͳ͘طଘͷࢿ࢈ Λ׆͔ͯ͠࡞Δ͜ͱ͕ग़དྷΔ • ෦ͷྻʹApache ArrowʹରԠͨ͠numo- narrayͤ͞Δ͜ͱͰApache Arrowܗࣜʹม͢Δ ͜ͱՄೳ 37
Slide 38
Slide 38 text
38 MNIST
Slide 39
Slide 39 text
Agenda • Rubyʹ͓͚Δػցֶशͷݱঢ়ʹ͍ͭͯ • Red Data Toolsʹ͍ͭͯ • ݱঢ়ͷऔΓΈ • ະདྷʹ͍ͭͯ • ·ͱΊ 39
Slide 40
Slide 40 text
Red Data Toolsͷࠓޙ • Ҿ͖ଓ͖طଘͷgemͷAapche ArrowͷରԠߦ ͏ • Red ChainerͷΑ͏ͳ৽͍͠πʔϧͷ࡞ 40 3VCZؒͰσʔλੳ͕ ग़དྷΔΑ͏ʹͳΓ͍ͨ
Slide 41
Slide 41 text
Red Data Toolsͷࠓޙ Apache Arrowຊମͷ։ൃͷࢀՃ 41 "QBDIF"SSPX $ FUD
Slide 42
Slide 42 text
Agenda • Rubyʹ͓͚Δػցֶशͷݱঢ়ʹ͍ͭͯ • Red Data Toolsʹ͍ͭͯ • ݱঢ়ͷऔΓΈ • ະདྷʹ͍ͭͯ • ·ͱΊ 42
Slide 43
Slide 43 text
·ͱΊ w "QBDIF"SSPXʹରԠ͢Δ͜ͱͰඞཁͳ෦͔Β 3VCZͰσʔλੳΛ͡ΊΔ͜ͱ͕ग़དྷΔ w গͮͭ͠Ͱ3VCZͰσʔλੳ͕ग़དྷΔະདྷʂ 43
Slide 44
Slide 44 text
44 ࠂ
Slide 45
Slide 45 text
։ൃΠϕϯτͬͯ·͢ 45 w ຖ݄ճͷఆظ։࠵ w ॴ4QFFFͰߦͬͯ·͢!౦ژຊ w σʔλੳ͜Ε͔ΒͷਓͰͲͳͨͰ0,Ͱ͢ʂ
Slide 46
Slide 46 text
Gitter 46 w ຊޠɹIUUQTHJUUFSJNSFEEBUBUPPMTKB w &OHMJTIIUUQTHJUUFSJNSFEEBUBUPPMTFO