ϝδϟʔϦʔΨʔͷ ࡱΕߴʢPythonฤʣ Shinichi Nakagawa(@shinyorke) kawasaki.rb #069 2019/2/27

Who am I? • Shinichi Nakagawa(@shinyorke) • ʢגʣωΫετϕʔε ໺ٿΤϯδχΞ/CTO • #rettypy ओ࠵ऀ • ΤϯδχΞ࠾༻ɾٕज़޿ใྺ3೥

ͦͷTechϒϩάຊ౰ʹඞཁͰ͔͢ʁ ʮ࠾༻޿ใʯΛޠΔ޿ใLTେձ#17@αΠϘ΢ζ #PRLT https://speakerdeck.com/shinyorke/sofalsetechburoguben-dang-nibi-yao-desuka-burogufalse-cuo-regao- number-prlt

ϒϩάͷࡱΕߴ=Ԡื਺ ※ձࣾͷٕज़ϒϩάͷ࿩Ͱ͢ʂ
 ʢݸਓ͸Ͳ͏͔Θ͔ΒΜʣ

໺ٿબखͷʮࡱΕߴʯ = ಘ఺Ձ஋ ೋྥଧҰຊ͸ͲΕ͙Β͍఺ʹͭͳ͕Δʁ ૹΓόϯτΛఆྔతʹධՁͬͯʁʁ

໺ٿબखͷʮࡱΕߴʯࢉग़ • શଧ੮ͷϓϨʔΛಘ఺΁ͷߩݙ౓ͱͯ͠ఆྔԽ
 ʮಘ఺ظ଴஋ʯͱݺ͹ΕΔࢦඪɾߟ͑ํͰ΍Δ • ଧ੮ʹཱͬͨ࣌ͷಘ఺ظ଴஋ͱɺ
 ଧ੮ऴྃޙͷಘ఺ظ଴஋ͷࠩ෼Ͱ
 ʮϓϨʔ͕ಘ఺ʹͭͳ͕͔ͬͨʁʯΛग़͢
 ˠಘ఺Ձ஋ͱݺ͹ΕΔ΋ͷ

ಘ఺ظ଴஋ͱಘ఺Ձ஋ʢৄ͘͠ʣ • ϥϯφʔͷ਺(8௨Γ)×Ξ΢τΧ΢ϯτ(3௨Γ)=24௨Γͷঢ়گΛ෼ ྨ,͔ͦ͜Β3Ξ΢τऔΒΕΔ·Ͱʹ֫ಘͰ͖Δ(ͱࢥΘΕΔ)ฏۉత ͳಘ఺Λʮಘ఺ظ଴஋(Run Expectancy)ʯͱݺͿ. • ϓϨʔ(ώοτ,૸ྥ,etc…)ʹΑͬͯ,ಘ఺ظ଴஋Λ্͔͛ͨ(·ͨ͸ Լ͔͛ͨ)ΛੵΈॏͶͯબखΛධՁ͢Δ.
 ͜ΕΛʮಘ఺Ձ஋(Run Value)ʯͱݺͿ. • Α͘Θ͔Μͳ͍ਓ͸ɺAnalyzing Baseball Data with R
 ΋͘͠͸ϚωʔɾϘʔϧΛಡΜͰ͍ͩ͘͞ʂ

PythonͰࢉग़ͯ͠ΈΔ • Analyzing Baseball Data with Rͱ͍͏ຊʹɺ
 RͰͷܭࢉํ๏͕͋ΔͷͰͦΕΛRͰࣸܦ • R͔ΒPythonʹॻ͖׵͑ • جຊతʹ͸pandasͷؔ਺͚ͩͰ࣮૷

ͪͳΈʹσʔλ͸ • ࠓ೥ͷϝδϟʔϦʔάͷશଧ੮σʔλ
 retrosheet͍ͬͯ͏ެ։σʔληοτ • CSVϑΝΠϧɺ110MBͪΐ͍ • ໿19ສߦɺ96ྻʢ࢖͏ͷ͸10ྻ΋ແ͍ʣ

ಘ఺ظ଴஋ʢMLB 2018೥ʣ த઒͕ಠࣗࢉग़, MLBͷαΠτͱಉ͡਺஋ͳͷͰਖ਼ղͷ͸ͣ ݩσʔλɿ https://github.com/chadwickbureau/baseballdatabank ແࢮ Ұࢮ ೋࢮ ϥϯφʔແ͠ 0.49 0.26 0.10 Ұྥ 0.87 0.53 0.22 ೋྥ 1.13 0.68 0.32 ࡾྥ 1.43 1.00 0.35 Ұྥೋྥ 1.42 0.93 0.44 Ұྥࡾྥ 1.79 1.21 0.50 ೋྥࡾྥ 1.94 1.36 0.57 ຬྥ 2.35 1.47 0.77

Run Value = New State - State + Run Scored Run valueɿಘ఺Ձ஋ʢࡱΕߴʣ New Stateɿଧ੮݁Ռͷಘ఺ظ଴஋ Stateɿଧ੮ʹཱͭલͷಘ఺ظ଴஋ Run Scoredɿ࣮ࡍʹೖͬͨಘ఺ʢ0ʙ4ʣ

ܭࢉྫ • ແࢮ1ྥ͔Β໨͕֮ΊΔೋྥଧͰແࢮ2,3ྥ
 1.94(2,3ྥ) - 0.87(1ྥ) + 0(఺) = 1.07
 ࡱΕߴ͋Δύλʔϯ • ແࢮ1ྥ͔ΒଉΛٵ͏༻ʹόϯτ੒ޭ1ࢮ2ྥ
 0.68(1ࢮ2ྥ) - 0.87(1ྥ) + 0(఺) = -0.19
 ΉΉʁΉ͠ΖԼ͕ͬͯΔͧʁʁ • ୯७ͳྫ͕ͩ͜ΕͰϓϨʔධՁՄೳ

͜ΕͰϝδϟʔϦʔΨʔʹ౰ͯ͸Ί ධՁ͢ΔͱͲ͏ͳΔ͔ʁʁʁ

…ͱ͍͏ଓ͖ͷ࿩͸ɺ ʮBaseball Play Study 2019य़ʯ Ͱ൸࿐͢Δʢ͔΋ʣ ※3/27(ਫ)ϓϨΠϘʔϧ⽁ #bpstudy

͓͠·͍.