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
広告配信サーバーと広告配信比率最適化問題
Search
Ken Wagatsuma
February 10, 2018
Technology
1
990
広告配信サーバーと広告配信比率最適化問題
Lightening Talk at
https://techconf.cookpad.com/2018/
Ken Wagatsuma
February 10, 2018
Tweet
Share
More Decks by Ken Wagatsuma
See All by Ken Wagatsuma
Pregel Graph Compute Engines - Supersteps Exampls
kenju
0
210
Kafka on Kubernetes with Strimzi
kenju
0
160
AWS DynamoDB Accelerator (DAX) 101
kenju
2
7.1k
Moden browser introduction
kenju
1
410
Cookpad summer internship 2019 - API
kenju
0
10k
Introduction to Design Patterns
kenju
0
82
GraphQL Asia 2019 "Re-architecture of a decade-old app with BFF/GraphQL"
kenju
0
8.9k
Introduction to TypeScript
kenju
0
720
Introduction to Programmatic Ad
kenju
0
260
Other Decks in Technology
See All in Technology
Power Automate のパフォーマンス改善レシピ / Power Automate Performance Improvement Recipes
karamem0
0
200
PHPでResult型やってみよう
higaki_program
0
190
本当にわかりやすいAIエージェント入門
segavvy
10
6k
激動の時代、新卒エンジニアはAIツールにどう向き合うか。 [LayerX Bet AI Day Countdown LT Day1 ツールの選択]
tak848
0
550
機械学習を「社会実装」するということ 2025年夏版 / Social Implementation of Machine Learning July 2025 Version
moepy_stats
1
750
データエンジニアリング 4年前と変わったこと、 4年前と変わらないこと
tanakarian
2
370
「手を動かした者だけが世界を変える」ソフトウェア開発だけではない開発者人生
onishi
14
6.1k
少人数でも回る! DevinとPlaybookで支える運用改善
ishikawa_pro
3
700
Railsの限界を超えろ!「家族アルバム みてね」の画像・動画の大規模アップロードを支えるアーキテクチャの変遷
ojima_h
3
450
AI工学特論: MLOps・継続的評価
asei
10
1.7k
(HackFes)米国国防総省のDevSecOpsライフサイクルをAWSのセキュリティサービスとOSSで実現
syoshie
5
660
DatabricksのOLTPデータベース『Lakebase』に詳しくなろう!
inoutk
0
110
Featured
See All Featured
個人開発の失敗を避けるイケてる考え方 / tips for indie hackers
panda_program
109
19k
Designing for Performance
lara
610
69k
Facilitating Awesome Meetings
lara
54
6.5k
Connecting the Dots Between Site Speed, User Experience & Your Business [WebExpo 2025]
tammyeverts
8
370
Imperfection Machines: The Place of Print at Facebook
scottboms
267
13k
How GitHub (no longer) Works
holman
314
140k
Navigating Team Friction
lara
187
15k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
21
1.3k
Being A Developer After 40
akosma
90
590k
Principles of Awesome APIs and How to Build Them.
keavy
126
17k
The Myth of the Modular Monolith - Day 2 Keynote - Rails World 2024
eileencodes
26
2.9k
For a Future-Friendly Web
brad_frost
179
9.8k
Transcript
ࠂ৴αʔόʔͱ ࠂ৴ൺ࠷దԽ ϝσΟΞϓϩμΫτ։ൃ෦ ,FOKV8BHBUTVNB
8IP Kenju Wagatsuma (github.com/kenju) • ϝσΟΞϓϩμΫτ։ൃ෦ • αʔόʔαΠυΤϯδχΞ • ͖ͳͷɿRuby,
ίʔώʔ, ϩδΧϧΫοΩϯά • ݏ͍ͳͷɿ1ϲ݄લʹॻ͍ͨࣗͷίʔυ
ϝσΟΞϓϩμΫτ։ൃ෦ ୲αʔϏεɿ ࠂ৴, storeTV, cookpadTV, OEM, ͦͷଞଟ ࢀߟɿ ։ൃऀϒϩάʰΫοΫύουͷࠂΤϯδχΞԿΛ ͍ͬͯΔͷ͔ʱ
ຊ͍ͨ͜͠ͱɻ ϝσΟΞϓϩμΫτ։ൃ෦Ͱ ͲΜͳϓϩδΣΫτΛ͍ͬͯΔͷ͔ʁ
νʔϜʹೖͬͯϲ݄ޙʹऔΓΜͩϓϩδΣΫτ ΫοΫύουͷࠂ৴αʔόʔʹ͓͚Δ ࠂ৴ൺͷࣗಈ࠷దԽϓϩδΣΫτɻ
ݫ͍͠εέδϡʔϧ • ϝσΟΞϓϩμΫτ։ൃ෦δϣΠϯ - 10݄த० • ͓खฒΈഈݟϓϩδΣΫτ - ~11݄த० •
৴࠷దԽτϥΠΞϧ - 12/4(݄) 10:00 - 12/11(݄) 10:00 ???
ղܾ͍ͨ͠՝ • ʑͷखӡ༻ʹΑΔνϡʔχϯά͕ඞཁ - => ࡞ۀ͕ൃੜ • ӡ༻ऀͷܦݧͱצʹཔͬͨνϡʔχϯά - =>
ҟಈ࣌ಋೖ࣌ͷίετ͕ߴա͗ • ࠷దͳࡏݿൺΛࣗಈͰௐͰ͖ͳ͍ - => ࠂܝग़ͷػձଛࣦ
Ͳ͏ղܾ͢Δ͔ • ࡏݿׂྔͱ࣮͔Β࠷దͳ৴ൺͷิ ਖ਼Λߦ͏ - ΠϯϓϨογϣϯϕʔε͔ΒΫϦοΫϕʔεͷ৴ - ΫϦοΫ༧ଌΛར༻ͨ͠ൺͷࣗಈ࠷దԽ - ϦΞϧλΠϜूܭσʔλΛ׆༻ͨ͠ΞʔΩςΫνϟ
‣ Lambda Architecture ʹ͓͚Δ Speed Layer
l4QFFE-BZFSzPO"84 • Kinesis, DynamoDB, Lambda Λ׆༻ͨ͠ Speed Layer (from Lambda
Architecture) • طଘͷετϦʔϜʹɺΫ ϦοΫܭࢉϨΠϠʔΛ Ճ͚ͨͩ͠ = ઌਓͷݞ ʹΔ
ৄ͍ͪ͜͠Β ࢀߟɿ ʰCookpad Tech Kitchen #9 ʙ1ߦͷϩάͷ͜͏ ଆʙ Λ։࠵͠·ͨ͠ʂʱ
ΫϦοΫ༧ଌ͍͠ʂʂʂ • ޯϒʔεςΟϯάܾఆʢGBDTʣΛ༻͍ͨࠂ͝ͱͷΫϦοΫ༧ଌ - Facebook https://code.facebook.com/posts/975025089299409/evaluating-boosted-decision-trees-for-billions-of-users - SmartNews https://speakerdeck.com/komiya_atsushi/gbdt-niyorukuritukulu-yu-ce-wogao-su-hua-sitai-number-oresikanaito-vol-dot-4 •
ଟόϯσΟοτͷҰछͰ͋ΔMortal Multi-Armed BanditsͷԠ༻ - Voyage Group http://techlog.voyagegroup.com/entry/2015/04/03/114547ɹ • Neural Networkͷ૯߹֨ಆٕʢ͕͢͞Googleʣ - Google http://www.eecs.tufts.edu/~dsculley/papers/ad-click-prediction.pdfɹ • ৴པͱ࣮ͷϩδεςΟοΫճؼʢୠܻ͕͠ԯϨϕϧʣ - Criteo http://olivier.chapelle.cc/pub/ngdstone.pdfɹ
ؒʹ߹Θͳ͍ʂ • τϥΠΞϧͳΜͱͯ͠ʹ࣮ࢪ͍ͨ͠ - վળͷαΠΫϧΛճͨ͢Ί • QCDͰݴ͏ͳΒɺDelivery, QualityΛ༏ઌ - ͳΜͱͯؒ͠ʹ߹Θ͍ͤͨʂ
• ࠷ॳ͔Βᘳͳਫ਼༧ଌ·ͣෆՄೳ - ػցֶशͰղܾ͠ͳͯ͘Α͍͔·ͣߟ͑Δ - ࢀߟɿʰࣄͰ͡ΊΔػցֶशʱ
ҠಈฏۉԞ͕ਂ͍ • SMA (Simple Moving Average) = ۙ N ݸͷॏΈ͚ͷͳ͍୯७ͳฏۉ
• WMA (Weighted Moving Average) = ΑΓ࠷ۙͷσʔλʹॏΈ͚ • EWMA (Exponentially Weighted Moving Average) = ࢦؔతʹॏΈ͚ • MMA (Modified Moving Average) = EWMAͷѥछ ଞʹTriangle MA, Sine Weighted MA, KZ Filtering,...etc ࢀߟɿhttps://en.wikipedia.org/wiki/Moving_average#Simple_moving_averag
աڈϩάΛݩʹΞϧΰϦζϜͷਫ਼Λੳ • Jupyter Notebook / Python - ࢀߟɿ։ൃऀϒϩάʰRailsΤϯ δχΞʹཱͭJupyter Notebook
ͱiRubyʱ • ൺֱͨ͠ΞϧΰϦζϜ - Total Average - Cumulative Average - Simple Moving Average (3 Hours) - Simple Moving Average (6 Hours)
τϥΠΞϧ݁Ռ • ิਖ਼ͷϩδοΫʹ՝ ͕ݟ͔ͭͬͨ ͷɺτϥΠΞϧͱ͠ ͯޭ
ظతνϡʔχϯά • Speed Layer ͷ࠶ઃܭɾຏ͖ࠐΈ - ετϦʔϜॲཧʹԊͬͨσʔλͷྲྀΕ • ෛ࠴ =
ະୡ ΛՃຯͨ͠ϩδοΫ - ୈҰ࣍τϥΠΞϧΛ͍ͬͯͳ͔ͬͨΒݟ͑ͳ͔ͬͨ՝ • ҠಈฏۉΞϧΰϦζϜͷվળ - Batch LayerͰΦϑϥΠϯͰܭࢉ&࠷ਫ਼͕ྑ͍ͷΛબ - Gem࡞ͬͨ https://github.com/kenju/moving_avg-ruby
தظͰ͍͖ͬͯ • ΫϦοΫ༧ଌਫ਼ͷߋͳΔ্ˍ৽نࠂ։ൃ - ػցֶशϨΠϠʔͷຊ൪ಋೖ • Lambda Architectureͷຏ͖ࠐΈ - ࢀߟɿ։ൃऀϒϩάʰαʔόʔϨεͳόοΫΞοϓγεςϜ
Λ AWS SAM Λ༻͍ͯγϡοͱߏங͢Δʱ • ࠂ৴αʔόʔࣗମͷѹతվળ - ։ൃج൫ͷڥඋ - ύϑΥʔϚϯε࠷దԽɺϨΨγʔίʔυͷվળ
ຖͷྉཧΛָ͠Έʹ͢Δ 5IBOLZPV