Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
フリーランスだらけの ML基盤開発 / ML Infra Development with F...
Search
Kengo Miyakawa
December 12, 2019
Technology
3
2.7k
フリーランスだらけの ML基盤開発 / ML Infra Development with Freelance
Kengo Miyakawa
December 12, 2019
Tweet
Share
Other Decks in Technology
See All in Technology
障害対応指揮の意思決定と情報共有における価値観 / Waroom Meetup #2
arthur1
5
480
初心者向けAWS Securityの勉強会mini Security-JAWSを9ヶ月ぐらい実施してきての近況
cmusudakeisuke
0
130
Lexical Analysis
shigashiyama
1
150
なぜ今 AI Agent なのか _近藤憲児
kenjikondobai
4
1.4k
iOS/Androidで同じUI体験をネ イティブで作成する際に気をつ けたい落とし穴
fumiyasac0921
1
110
Application Development WG Intro at AppDeveloperCon
salaboy
0
190
OCI Network Firewall 概要
oracle4engineer
PRO
0
4.1k
VideoMamba: State Space Model for Efficient Video Understanding
chou500
0
190
Flutterによる 効率的なAndroid・iOS・Webアプリケーション開発の事例
recruitengineers
PRO
0
110
Lambda10周年!Lambdaは何をもたらしたか
smt7174
2
110
Amazon CloudWatch Network Monitor のススメ
yuki_ink
1
210
Exadata Database Service on Dedicated Infrastructure(ExaDB-D) UI スクリーン・キャプチャ集
oracle4engineer
PRO
2
3.2k
Featured
See All Featured
How to train your dragon (web standard)
notwaldorf
88
5.7k
Building Your Own Lightsaber
phodgson
103
6.1k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
280
13k
Six Lessons from altMBA
skipperchong
27
3.5k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
191
16k
StorybookのUI Testing Handbookを読んだ
zakiyama
27
5.3k
Writing Fast Ruby
sferik
627
61k
Building an army of robots
kneath
302
43k
How to Think Like a Performance Engineer
csswizardry
20
1.1k
Designing for Performance
lara
604
68k
Helping Users Find Their Own Way: Creating Modern Search Experiences
danielanewman
29
2.3k
Fashionably flexible responsive web design (full day workshop)
malarkey
405
65k
Transcript
ϑϦʔϥϯεͩΒ͚ͷ MLج൫։ൃ ٶ݈ޗ @MLPP#5 2019/12/12 1
ٶ ݈ޗ (@N30nnnn) ࣗݾհ • ܦӦֶ - ౷ܭֶઐ߈ • ݸਓࣄۀओ
• σʔλੳ → ը૾ܥML → EM + MLOps • ݱࡏ2ࣾͰ MLOps 2
৬ྺ • ࠂཧళ (σʔλੳ) • ਓೳडୗ։ൃ1 (ML) • ਓೳडୗ։ൃ2 (ML,
EM) • (ݱ৬) גࣜձࣾΫϥϏε (MLOps, EM) ᵋ ձܭܥը૾ॲཧ • (ݱ৬) גࣜձࣾϨΞδϣϒ (MLOps) ᵋ ΦϯϥΠϯӳձ EM: HowͱWho EM: HowͱWho 3 ※ڐՄΛड͚ͯެ։ ※2019/12 ݱࡏ
MLνʔϜͷϝϯόʔߏ • (ݱ৬) גࣜձࣾΫϥϏε (MLOps, EM) ᵋ ձܭܥը૾ॲཧ • (ݱ৬)
גࣜձࣾϨΞδϣϒ (MLOps) ᵋ ΦϯϥΠϯӳձ ਖ਼ࣾһ: 2ਓ ϑϦʔ: 7ਓ ਖ਼ࣾһ: 2ਓ ϑϦʔ: 5ਓ 4 ※ڐՄΛड͚ͯެ։ ※2019/12 ݱࡏ
Agenda • ࠷ۙ࡞͍ͬͯΔϓϩμΫτͷߏ • Fargateϕʔε • EKSϕʔε • ϑϦʔϥϯε͕ଟ͍ϓϩδΣΫτͷΈ •
ਓࡐαΠΫϧͷ͞ • ઐྖҬͷҧ͍ • ͞ΕͨMLίʔυཧ • ߨͨ͡ղܾࡦ • ΠϯλʔϑΣʔεͷపఈ - ؔ৺ͷ 5
Fargateͷࣄྫ 6
ϓϩμΫτ: Fargate - ਪڥ • 1API - 1ϦϙδτϦ • gitflowͰ
develop/master → ։ൃ/ຊ൪ • ΠϝʔδϏϧυͱ devσϓϩΠCircleCIͰࣗಈԽ • ΠϯϑϥTerraform 7 ※ڐՄΛड͚ͯެ։
ϓϩμΫτ: Fargate - ࠶ֶशػߏ 8 • ֶशσʔλRDSʹ. • ఆظతʹֶशΛτϦΨ. •
Dynamoʹֶश݁Ռอଘ • Dynamoࢀর͠࠷ྑͷϞσ ϧΛऔಘ • RDSʹཷΊΔͷΈͰɺ ੑೳ্/มԽΛଊ͑ ͍ͨ ※ڐՄΛड͚ͯެ։
ϓϩμΫτ: Fargate - ࠶ֶशػߏ 9 • ֶशσʔλRDSʹ. • ఆظతʹֶशΛτϦΨ. •
Dynamoʹֶश݁Ռอଘ • Dynamoࢀর͠࠷ྑͷϞσ ϧΛऔಘ • RDSʹཷΊΔͷΈͰɺ ੑೳ্/มԽΛଊ͑ ͍ͨ ※ڐՄΛड͚ͯެ։
ϓϩμΫτ: Fargate - ࠶ֶशػߏ 10 • ֶशσʔλRDSʹ. • ఆظతʹֶशΛτϦΨ. •
Dynamoʹֶश݁Ռอଘ • Dynamoࢀর͠࠷ྑͷϞσ ϧΛऔಘ • RDSʹཷΊΔͷΈͰɺ ੑೳ্/มԽΛଊ͑ ͍ͨ ※ڐՄΛड͚ͯެ։
ϓϩμΫτ: Fargate - ࠶ֶशػߏ 11 • ֶशσʔλRDSʹ. • ఆظతʹֶशΛτϦΨ. •
Dynamoʹֶश݁Ռอଘ • Dynamoࢀর͠࠷ྑͷϞσ ϧΛऔಘ • RDSʹཷΊΔͷΈͰɺ ੑೳ্/มԽΛଊ͑ ͍ͨ ※ڐՄΛड͚ͯެ։
EKSͷࣄྫ 12
ϓϩμΫτ: EKS - ਪڥ • ෳͷೖྗΛ·ͱΊͯਪఆ ΩϡʔܗࣜͷAPI • ࠶ֶशݕ౼த •
develop/master → ։ൃ/ຊ൪ • ΠϝʔδϏϧυ CircleCIͰࣗಈԽ • ΠϯϑϥTerraform 13 ※ڐՄΛड͚ͯެ։
ϓϩμΫτ: EKS - ਪڥ • ෳͷೖྗΛ·ͱΊͯਪఆ ΩϡʔܗࣜͷAPI • ࠶ֶशݕ౼த •
develop/master → ։ൃ/ຊ൪ • ΠϝʔδϏϧυ CircleCIͰࣗಈԽ • ΠϯϑϥTerraform 14 ※ڐՄΛड͚ͯެ։
ϓϩμΫτ: EKS - ਪڥ • ෳͷೖྗΛ·ͱΊͯਪఆ ΩϡʔܗࣜͷAPI • ࠶ֶशݕ౼த •
develop/master → ։ൃ/ຊ൪ • ΠϝʔδϏϧυ CircleCIͰࣗಈԽ • ΠϯϑϥTerraform 15 ※ڐՄΛड͚ͯެ։
ϓϩμΫτ: EKS - ਪڥ • ෳͷೖྗΛ·ͱΊͯਪఆ ΩϡʔܗࣜͷAPI • ࠶ֶशݕ౼த •
develop/master → ։ൃ/ຊ൪ • ΠϝʔδϏϧυ CircleCIͰࣗಈԽ • ΠϯϑϥTerraform 16 ※ڐՄΛड͚ͯެ։
• ͳΔ͘γϯϓϧͳ࡞Γ • δϣΠϯίετݮ • ຊ࣭తͳ։ൃʹઐ೦ • ҰํͰͳΔ͘Ϟμϯʹ • ৽͍ؒ͠ʹདྷͯͨ͘Ίͷڥ࡞Γ
17
ϑϦʔϥϯε͕ଟ͍PJͷΈ 18 • ਓࡐαΠΫϧͷ͞ • ઐྖҬͷҧ͍ • ͞ΕͨMLίʔυཧ
ϑϦʔϥϯε͕ଟ͍PJͷΈ1 • ਓࡐαΠΫϧͷ͞ • 3ϲ݄, 6ϲ݄Ͱ͍ͳ͘ͳΔ͜ͱ • ࠾༻ଆ߹ɾΤϯδχΞଆ߹ ΩϟονΞοϓͷ͕࣌ؒπϥ͍ 19
ϑϦʔϥϯε͕ଟ͍PJͷΈ2 • ઐྖҬͷҧ͍ • ඞͣ͠ML / αʔόʔαΠυ྆ํͷ͕ࣝ༗ΔΘ͚Ͱͳ͍ • αʔόʔαΠυͷ͠͞(e.g. APIपΓ,
ฒߦॲཧ) • MLͷ͠͞(e.g. ੑೳࢦඪ, Train,Validation,Test…) • ඇઐྖҬʹੵۃతͱݶΒͳ͍ ίϛϡχέʔγϣϯίετ͕πϥ͍ 20
ϑϦʔϥϯε͕ଟ͍PJͷΈ3 • ͞ΕͨMLίʔυཧ • લఏͱ͞ΕΔσΟϨΫτϦߏ / ਖ਼ղϥϕϧܗࣜ • ୯ҰͷࣈԽͮ͠Β͍ੑೳධՁ MLͷཧ͕πϥ͍
21
ΠϯλʔϑΣʔεͷపఈ 22
ؔ৺ͷ • Pythonʹଘࡏ͠ͳ͍ΠϯλʔϑΣʔε • σʔλͷࡏΓॲμϯϩʔυͳͲʹؔ༩ͨ͘͠ͳ͍ML • Ϟσϧͷൺֱํ๏લॲཧޙॲཧͳͲʹؔ༩ͨ͘͠ͳ͍αʔό • ͞Εͨίʔυͷ࣮ଶѲΛ؆ૉԽ͍ͨ͠ཧऀ ΠϯλʔϑΣʔεΛݻΊͯίϛϡχέʔγϣϯΛݮΒ͢
23
e.g. ਪఆॲཧͷ߹ 24 αʔόʔαΠυ ML
e.g. ਪఆॲཧͷ߹ 25 αʔόʔαΠυ σʔλͷॴࡏ, औಘํ๏ΛML͔Β ฦ٫࣌ͷܗΛ αʔό͔Β ML σʔλͷॴࡏΛML͕
ؾʹ͢Δඞཁ͕ͳ͍
e.g. ࠶ֶशػߏ αʔόʔαΠυ ML αʔόʔαΠυ 26 ݺͼग़͠
e.g. ࠶ֶशػߏ αʔόʔαΠυ ML αʔόʔαΠυ σʔλͷॴࡏ, औಘํ๏ΛML͔Β σʔλͷॴࡏ, औಘํ๏ΛML͔Β ੑೳൺֱͷ
Ϟσϧબͷ໌ࣔ 27 Ϟσϧաڈใͷॴࡏ, औಘ,هΛ Ϟσϧաڈใͷॴࡏ, औಘ,هΛ
e.g. ࠶ֶशػߏ αʔόʔαΠυ ML αʔόʔαΠυ ࠶ֶश༻σʔλϕʔε •ը૾ྨςʔϒϧ •จࣈྻݕςʔϒϧ •OCRςʔϒϧ •……
நԽ 28
• ݸʑਓͷઐྖҬੌ͍ • ϑϦʔϥϯεͷੜ໋ઢ • ؔ৺֎ʹؔ༩ͯ͠ΒΘͳͯ͘ྑ͍ߏ • ҟͳΔઐؒͰͷίϛϡχέʔγϣϯݮ͠ෛ୲ݮ • αʔόʔαΠυͷΈͳΒͣMLϑϦʔϥϯεͰՄೳʹ
29 ·ͱΊ