- Prophetアルゴリズムの基本 - 検証のやり方 - 季節性での加法と乗法の使い分け - 変数重要度と変数ごとの効果
Exploratory ηϛφʔ #25ProphetΛͬͨ࣌ܥྻ༧ଌ
View Slide
EXPLORATORY
3εϐʔΧʔా צҰCEOEXPLORATORYུྺ2016य़ɺσʔλαΠΤϯεͷຽओԽͷͨΊɺExploratory, Inc Λ্ཱͪ͛ΔɻExploratory, Inc.ͰCEOΛΊΔ͔ͨΘΒɺσʔλαΠΤϯεɾϒʔτΩϟϯϓɾτϨʔχϯάͳͲΛ௨ͯ͠σʔλαΠΤϯεͷٕज़ͱख๏ͷීٴͱڭҭʹऔΓΉɻถΦϥΫϧຊࣾͰɺ16ʹΘͨΓσʔλαΠΤϯεͷ։ൃνʔϜΛ͍ɺػցֶशɺϏοάɾσʔλɺϏδωεɾΠϯςϦδΣϯεɺσʔλϕʔεʹؔ͢Δଟ͘ͷΛੈʹૹΓग़ͨ͠ɻ@KanAugust
VisionΑΓΑ͍ҙࢥܾఆΛ͢ΔͨΊʹσʔλΛ͏͜ͱ͕ͨΓલʹͳΔ
MissionσʔλαΠΤϯεͷຽओԽ
6ୈ̏ͷσʔλαΠΤϯεɺAIɺػցֶश౷ܭֶऀɺ։ൃऀͷͨΊ͚ͩͷͷͰ͋Γ·ͤΜɻσʔλʹڵຯͷ͋ΔਓͳΒ୭͕ੈքͰ࠷ઌͷΞϧΰϦζϜΛͬͯϏδωεσʔλΛ؆୯ʹੳͰ͖Δ͖Ͱ͢ɻExploratory͕ͦ͏ͨ͠ੈքΛՄೳʹ͠·͢ɻ
ୈ1ͷ ୈ̎ͷ ୈ̏ͷϓϥΠϕʔτ(ߴ͍/ݹ͍) Φʔϓϯɾιʔε(ແྉ/࠷ઌ)UI & ϓϩάϥϛϯά ϓϩάϥϛϯά201620001976ϚωλΠθʔγϣϯ ίϞσΟςΟԽ ຽओԽ౷ܭֶऀ σʔλαΠΤϯςΟετExploratoryΞϧΰϦζϜϢʔβʔɾମݧπʔϧΦʔϓϯɾιʔε(ແྉ/࠷ઌ)UI & ࣗಈԽϏδωεɾϢʔβʔςʔϚσʔλαΠΤϯεͷຽओԽ
質問ExploratoryɹϞμϯˍγϯϓϧ UI伝えるデータアクセスデータラングリング可視化アナリティクス統計/機械学習
࣌ܥྻ༧ଌϞσϧ• ڭࢣ͋ΓֶशͷҰछɻ• աڈͷ࣌ܥྻσʔλΛτϨʔχϯάσʔλͱͯ͠ɺকདྷͷΛ༧ଌ͢ΔϞσϧΛ࡞Δɻ10
11աڈNͷσʔλ͔Βɺ࣍ͷͷΛ༧ଌ͢ΔϞσϧΛ࡞͢Δɻ࣌ܥྻ༧ଌ - ౷తͳΞϓϩʔνDay 1 Day 2 Day 3 Day 4Day 2 Day 3 Day 4 Day 5Day 3 Day 4 Day 5 Day 6
12• σʔλؒͷִ࣌ؒؒσʔλΛ௨ͯ͠ҰఆͰ͋Δඞཁ͕͋Δɻ• ͕NA (ܽଛ)ͱͳΔ͕͋ͬͯͳΒͳ͍ɻ• ෳͷपظੑ (िͱ) Λಉ࣌ʹѻ͏ͷ͍͠ɻ• ύϥϝʔλͷઃఆʹɺઐతͳ͕ࣝඞཁɻ౷తͳΞϓϩʔνͰͷ࣌ܥྻ༧ଌͷ
13• Facebookʹ͍ͨσʔλαΠΤϯςΟετʢSean J. Taylor & co.ʣ͕࡞ͬͨ࣌ܥྻ༧ଌΞϧΰϦζϜͰɺΦʔϓϯιʔεͱͯ͠ެ։͞Ε͍ͯΔɻ(https://facebook.github.io/prophet)• ౷ܭɺ࣌ܥྻ༧ଌͷઐ͕ࣝͳͯ͑͘ΔΑ͏ʹσβΠϯ͞Ε͍ͯΔɻProphetSean J. Taylor@seanjtaylor
14• ҎԼͷཁૉͷͱͯ͠ද͢͜ͱͷͰ͖ΔΒ͔ͳۂઢͷ͏ͪɺաڈσʔλʹ࠷ϑΟοτ͢ΔͷΛ୳͢ɻͦͷۂઢΛະདྷʹԆ͢Δ͜ͱʹΑͬͯ༧ଌ͢Δɻ• େہతͳτϨϯυ• पظͷقઅੑ (ɺिɺͳͲ)• ॕޮՌ - ΫϦεϚεɺ৽ɺ࣍ΠϕϯτɺͳͲɻ• ֎෦༧ଌมProphet
15େہతͳτϨϯυ
16େہతͳτϨϯυ
17पظͷقઅੑ
18τϨϯυ + पظੑ
19िपظͷมಈ
20τϨϯυ + पظੑ + िपظੑ
21τϨϯυ + पظੑ + िपظੑ21
22ॕޮՌ
23τϨϯυ + पظੑ + िपظੑ + ॕޮՌ
24• ࣌ؒͷൃలΛϞσϧͰදݱ͢Δ͜ͱ͖͋ΒΊΔɻ• ͔ΘΓʹɺ୯ʹۂઢΛݟ͚ͭΔͱ͍͏ʹ͢Δ͜ͱʹΑͬͯҎԼͷΑ͏ͳརΛಘ͍ͯΔɻ• σʔλؒͷִ͕࣌ؒؒҰఆͰ͋Δඞཁͳ͍ɻ• ͕NA (ܽଛ)ͱͳΔ͕͋ͬͯͳ͍ɻ• ෳͷपظੑ (िͱ) ͕σϑΥϧτͰߟྀ͞ΕΔɻ• σϑΥϧτͷઃఆͰͦΕͳΓͷ༧ଌ͕Ͱ͖ΔɻઃఆՄೳͳύϥϝʔλͷଟ͘ઐࣝແ͠ͰཧղՄೳɻProphetͷར
Let’s do it!25
Global Sales σʔλ26
27Sales (ച্)ΛϥΠϯνϟʔτͰՄࢹԽͯ͠ΈΔ• X࣠ʹOrder DateΛɺ࣌ؒ୯ҐΛWEEKͱׂͯ͠ΓͯΔɻ• Y࣠ʹSalesΛɺؔΛSUM(߹ܭ)ͱׂͯ͠ΓͯΔɻ
28
Salesʢച্ʣΛ༧ଌ͢ΔϞσϧΛ࡞Δ29
ΞφϦςΟΫεɾϏϡʔΛબ͠ɺ࣌ܥྻ༧ଌΛબɻ30
• /࣌ؒͷྻʹOrder DateΛ࣌ؒ୯ҐΛWEEKͱׂͯ͠ΓͯΔɻ• ͷྻʹSalesΛɺؔΛSUM(߹ܭ)ͱׂͯ͠ΓͯΔɻ31
ϓϩύςΟɾμΠΞϩάΛͬͯɺ1ؒͷ༧ଌΛ͢ΔͨΊʹɺ༧ଌظؒΛ52 (1ʹ52ि) ʹઃఆ͢Δɻ32
࣮ߦϘλϯΛΫϦοΫ͢Δͱ༧ଌ͞Εͨσʔλ͕ΦϨϯδ৭ͷϥΠϯͰදࣔ͞ΕΔɻ33ΦϨϯδ৭ͷઢ͚ͩͷ۠ؒࠓޙ1ؒͷ༧ଌΛද͢ɻ ୶͍ΦϨϯδ৭༧ଌͷෆ֬ఆ۠ؒ(uncertainty interval)Λද͢ɻ
34τϨϯυ λϒΛΫϦοΫ͢ΔͱτϨϯυϥΠϯͷೖͬͨνϟʔτ͕දࣔ͞ΕΔάϦʔϯͷઢ͕τϨϯυϥΠϯ
35ͷόʔɺτϨϯυʹมԽ͕͋ͬͨ࣌ʢνΣϯδϙΠϯτʣΛද͢ɻͷߴ͞ɺνΣϯδϙΠϯτͰͷτϨϯυϥΠϯͷ͖ͷมԽྔɻ
36पظλϒΛΫϦοΫ͢Δͱपظͷνϟʔτ͕දࣔ͞ΕΔɻ
Every year, the sales doesn’t pick up until June, then it goes down in July.
Weekly tab shows up only when the data is daily or more granular levels (hour, minutes, etc.)
Every week, the sales are low on Sunday and Monday,and the rest of the week is high.
40िपظΛݟΔͨΊʹɺ࣌ؒ୯ҐΛʹมߋ࣮ͯ͠ߦ͢͠ɻ
σʔλͷલॲཧܽଛʢNAʣͷॲཧ
Somehow, the weekly seasonality doesn’t repeat exactly the same…
There are NA for some dates. You can impute NA as part of the Data Preprocessing.
45• จ͕Ұ݅ͳ͍ͷച্Λ0ͱͯ͠ѻ͏ͨΊɺͷྻʹ͕ͳ͍ͱ͖ͷॲཧʹ”θϩͰຒΊΔ”ΛબͿɻ• ͜ΕΛ͠ͳ͍߹ɺจ͕Ұ݅ͳ͍ɺσʔλ͕ແ͍ͱͯ͠ѻΘΕΔɻ
Under the Importance tab, you can see which seasonality has more effecton the forecasting outcome.
49िपظλϒΛΫϦοΫ͢Δͱिपظͷνϟʔτ͕දࣔ͞ΕΔɻ
50ޮՌλϒΛΫϦοΫ͢Δͱɺ༧ଌΛߏ͢Δ֤ཁૉʢτϨϯυɺقઅੑʣ͕දࣔ͞ΕΔɻ
51มॏཁλϒΛΫϦοΫ͢ΔͱԿ͕༧ଌʹେ͖ͳӨڹΛ༩͍͑ͯΔͷ͔͕ݟ͑Δɻ
52σʔλλϒΛΫϦοΫ͢Δͱ༧ଌ͖ͷσʔλ͕දࣔ͞ΕΔ
• forecasted_value - ༧ଌ• forecasted_value_high/forecasted_value_low - ෆ֬ఆ۠ؒ• trend - େہతͳτϨϯυ• yearly - पظͷτϨϯυ• weekly - िपظͷτϨϯυ53༧ଌ͖ͷσʔλͷಡΈํ
54࣌ܥྻ༧ଌͷධՁ
55όοΫςετ• աڈσʔλͷ͏ͪɺ৽͍͠ظؒΛςετ༻ʹͱ͓ͬͯ͘ɻ• ςετظؒͷσʔλΛɺͦΕΑΓҎલͷσʔλΛͬͯ༧ଌ͢Δ• ༧ଌσʔλͱɺ࣮ࡍͷςετظؒͷσʔλΛൺֱධՁ͢Δɻ࣌ܥྻ༧ଌͷςετ๏
σʔλΛ2ͭͷηΫγϣϯʹ͚Δɻ56τϨʔχϯάظؒ ςετظؒ
57τϨʔχϯάσʔλΛͬͯ༧ଌϞσϧΛ࡞ΓɺςετظؒΛ༧ଌɻςετظؒͷ࣮ଌͱ༧ଌͷʮͣΕʯͷେ͖͞ΛධՁ͢ΔɻτϨʔχϯάظؒ ςετظؒ
58• ࣌ؒͷ୯ҐʹMON(݄)Λࢦఆ͢ΔɻςετϞʔυʹΓସ͑Δ
59ςετϞʔυΛTRUEʹ͠ɺςετظؒΛ12 (݄) ͱ͢Δɻ
ࠨଆͷ੨͍ઢτϨʔχϯάσʔλɺӈଆͷਫ৭ͷઢςετσʔλɻ60
61੨৭ͷઢ͕τϨʔχϯάσʔλɻ͜ͷൣғσʔλΛͱʹɺ༧ଌϞσϧ͕࡞ΒΕΔɻ
ΦϨϯδͷઢ͕ɺτϨʔχϯάσʔλΛݩʹ࡞ΒΕͨϞσϧʹΑΔ༧ଌσʔλɺബ͍ΦϨϯδͷྖҬͦͷ৴པ۠ؒͰ͋Δɻ62
͜ͷൣғͷσʔλΛͱʹ༧ଌϞσϧ͕࡞ΒΕ͍ͯΔͨΊɺ ͜ͷൣғͷ࣮σʔλͱϞσϧʹΑΔ༧ଌσʔλ͔ͳΓҰக͍ͯ͠Δɻ63
ςετظؒͷ༧ଌσʔλͱ࣮ଌͷζϨ͕ͲΕ͚ͩେ͖͍ͷ͔Λࢉग़͠ɺͦΕΛ࣋ͬͯϞσϧͷਫ਼ΛධՁ͍ͨ͠ɻ64
ςετ݁ՌͷαϚϦ65
66• RMSE (Root Mean Square Error) : ༧ଌ͔ΒͷͣΕͷೋͷฏۉͷϧʔτ• MAE (Mean Absolute Error) : ༧ଌ͔ΒͷͣΕͷઈରͷฏۉ• MAPE (Mean Absolute Percentage Error) : ύʔηϯτͰදͨ͠༧ଌ͔ΒͷͣΕͷઈରͷฏۉ• MASE (Mean Absolute Scaled Error) : MAEΛɺτϨʔχϯάσʔλͰͷφΠʔϒ༧ଌʢҰͭલͷظͱಉ͕͡ݱΕΔͰ͋Ζ͏ͱ͍͏୯७ͳ༧ଌʣͷMAEͰׂͬͨͷɻ࣌ܥྻ༧ଌͷධՁࢦඪ
Rootʢฏํࠜʣ Meanʢฏۉʣ Squareʢ2ʣ Errorʢޡࠩʣͭ·Γɺ࣮ଌͱ༧ଌͷޡࠩΛ2ͯ͠ɺͦͷฏۉΛͱΓɺͦͷͷฏํࠜΛͱͬͨͷ͜ͱɻ67RMSE (Root Mean Square Error)
22 + 22 + 22 + 42 4 (ͷ)4 + 4 + 4 + 16 47 = 2.6568RMSE (Root Mean Square Error)2 2 42==ྫ͑ɺ࣮ଌͱ༧ଌͷޡ͕ࠩͦΕͧΕ2, 2, 2, 4ͩͬͨͱ͢ΔͱɺܭࢉҎԼͷΑ͏ʹͳΔɻ
Meanʢฏۉʣ Absoluteʢઈରʣ Errorʢޡࠩʣͭ·Γɺ࣮ଌͱ༧ଌͷޡࠩͷઈରͷฏۉ͜ͱɻ69MAE (Mean Absolute Error)
2 + 2 + 2 + 4 4 (ͷ)70ྫ͑ɺ࣮ଌͱ༧ଌͷޡ͕ࠩͦΕͧΕ2, 2, 2, 4ͩͬͨͱ͢ΔͱɺܭࢉҎԼͷΑ͏ʹͳΔɻ= 2.5MAE (Mean Absolute Error)2 2 42
Meanʢฏۉʣ Absoluteʢઈରʣ Percentageʢׂ߹ʣ Errorʢޡࠩʣͭ·Γɺ࣮ଌͱ༧ଌͷޡࠩͷׂ߹ͷઈରͷฏۉ͜ͱɻ71MAPE (Mean Absolute Percentage Error)
7212 13 1611·ͣɺ࣮ଌΛͱΊΔɻMAPE (Mean Absolute Percentage Error)
7312 13 1611MAPE (Mean Absolute Percentage Error)2 242࣍ʹɺ࣮ଌͱ༧ଌͷޡࠩΛͱΊΔɻ
74100 100 100100MAPE (Mean Absolute Percentage Error)16.6% 15.4%25%18.2%࣮ଌͱ༧ଌͷޡࠩΛ࣮ଌͰׂͬͯ100Λ͔͚ɺͦΕͧΕͷύʔηϯςʔδΛͱΊΔɻࣈ͕ϚΠφεͷ߹ɺϚΠφεͷූ߸ΛͱΔ (ઈର).
75100 100 100100MAPE (Mean Absolute Percentage Error)16.6% 15.4%25%18.2% 16.6 + 15.4 + 18.2 + 25 4 (ͷ)࠷ޙʹɺ͜ΕΒͷͷฏۉΛग़͢ɻ= 18.8%
76Meanʢฏۉʣ Absoluteʢઈରʣ ScaledʢεέʔϧௐࡁΈͷʣ ErrorʢޡࠩʣMAEΛҟͳΔεέʔϧͷσʔλͰͷ༧ଌͲ͏͠ͰൺֱՄೳͳΑ͏ʹεέʔϧௐͨ͠ͷɻMASE (Mean Absolute Scaled Error)
77MASE = ςετظؒͷMAE / τϨʔχϯάظؒͷφΠʔϒ༧ଌͷMAEMASEςετظؒͷMAEτϨʔχϯάظؒͷφΠʔϒ༧ଌͷMAE
78Ұظલͷ͕ɺࠓظͦͷ··ग़ΔͩΖ͏ɺͱ͍͏҆қͳ༧ଌɻφΠʔϒ༧ଌ
79φΠʔϒ༧ଌΛτϨʔχϯάظؒʹରͯ͠ߦͬͨͱ͖ͷMAEΛج४ͱͯ͠࠾༻MASEͷεέʔϧௐͷج४
80MASE = ςετظؒͷMAE / τϨʔχϯάظؒͷφΠʔϒ༧ଌͷMAEMASEςετظؒͷMAEτϨʔχϯάظؒͷφΠʔϒ༧ଌͷMAE
81قઅੑͷϞʔυՃ๏త vs ๏త
82The difference between the actual line and the forecasted linebecomes wider as the time progresses.
Կ͕ى͖͍ͯΔͷ͔ʁ83• ച্͕͢ΔʹͭΕɺͦΕʹ͋ΘͤͯقઅੑʹΑΔมಈେ͖͘ͳΔͱߟ͑Δͷ͕ࣗવɻ• ͔͠͠ϞσϧقઅੑʹΑΔมಈͷେ͖͍ͭ͞ͰҰఆͰ͋Δͱ͍͏લఏͰ༧ଌ͍ͯ͠Δɻ• ͕࣌ؒͨͬͯച্͕ͨ͋͠ͱͰɺϞσϧ͕༧ଌ͢Δقઅมಈͷେ͖͕͞ɺ࣮ࡍͷͷقઅมಈͷେ͖͞ʹ͍͍͚͍ͭͯͯͳ͍ͷͰͳ͍͔ʁ
ͱͯ͠Γཱ͍ͬͯΔͷϦχΞͳ৳ͼํ ੵͱͯ͠Γཱ͍ͬͯΔͷෳརޮՌͷ͋Δ৳ͼํམͪண͍ͯΔձࣾͷैۀһͷਪҠ Amazonͷച্ߴͷਪҠ͢Δͱ͖ͷͷ৳ͼํͷҧ͍84
ઌఔͷςετϞʔυͰͷ༧ଌ݁ՌΛ͏ҰݟͯΈΔͱ…85࣮ࡍͷقઅมಈɺ༧ଌ݁ՌΑΓେ͖͘ͳ͍ͬͯΔΑ͏ͩɻ
Կ͕ى͖͍ͯΔͷ͔ʁ86ച্ͷʹͱͳͬͯେ͖͘ͳΔقઅมಈΛϞσϧԽͰ͖ΕΑ͍ͷͰʁ• ച্͕͢ΔʹͭΕɺͦΕʹ͋ΘͤͯقઅੑʹΑΔมಈେ͖͘ͳΔͱߟ͑Δͷ͕ࣗવɻ• ͔͠͠ϞσϧقઅੑʹΑΔมಈͷେ͖͍ͭ͞ͰҰఆͰ͋Δͱ͍͏લఏͰ༧ଌ͍ͯ͠Δɻ• ͕࣌ؒͨͬͯച্͕ͨ͋͠ͱͰɺϞσϧ͕༧ଌ͢Δقઅมಈͷେ͖͕͞ɺ࣮ࡍͷͷقઅมಈͷେ͖͞ʹ͍͍͚͍ͭͯͯͳ͍ͷͰͳ͍͔ʁ
قઅੑϞʔυ87Ճ๏త͠ࢉͰޮՌ͕ݱΕΔɻ ྫɿ12݄ϓϥε$100,000 ͱͷʹؔΘΒͣมಈͷେ͖͞Ұఆɻ๏తֻ͚ࢉͰޮՌ͕ݱΕΔ ྫɿ12݄ϓϥε10% ͱͷ͕େ͖͚Εมಈେ͖͘ͳΔɻ
88Ճ๏త๏త
89Ճ๏త๏తقઅੑͷେ͖͞Ұఆ
90Ճ๏త๏తقઅੑͷେ͖͞ҰఆقઅੑݩͷʢτϨϯυʣʹൺྫͯ͠େ͖͘ͳΔ
91Ճ๏త๏తقઅੑͷେ͖͞ҰఆقઅੑݩͷʢτϨϯυʣʹൺྫͯ͠େ͖͘ͳΔ
قઅੑͷϞʔυʢՃ๏తɺ๏తʣͷΓସ͑92ΞφϦςΟΫεɾϓϩύςΟͷقઅੑͷϞʔυͰɺՃ๏త͔๏త͔ΛΓ͔͑Δ͜ͱ͕Ͱ͖Δɻ
93Ճ๏త๏త
94Ճ๏త๏త๏తͳقઅੑΛͬͨ༧ଌͷ΄͏͕࣮ଌʹ༧ଌ͕ਵ͍ͯ͠Δɻ
95Ճ๏త๏తςετ݁ՌΛΈΔͱɺ๏తͳقઅੑΛͬͨ༧ଌͷ΄͏͕ɺશͯͷࢦඪͰɺΑ͍݁Ռʢখ͍͞ʣʹͳ͍ͬͯΔɻ
قઅੑͷϞʔυʢՃ๏తɺ๏తʣͷൺֱ96قઅੑͷύλʔϯ͕ݟ͍͢Α͏ʹɺ࣌ؒͷ୯ҐΛ͍ͬͨΜWEEK(ि)ʹͯ͠ɺقઅੑͷϞʔυΛՃ๏తͱ๏తͷؒͰΓସ͑ͯɺ༧ଌ݁ՌΛൺֱͯ͠ΈΔɻ
݄୯Ґʹͯ͠ɺςετ݁ՌΛൺֱ97࣌ؒͷ୯ҐΛݩͷMON(݄)ʹͯ͠ɺقઅੑͷϞʔυΛՃ๏తͱ๏తͷؒͰΓସ͑ͯɺςετ݁ՌΛൺֱͯ͠ΈΔɻ
98ʮ܁Γฦ͠ʯΛͬͯෳͷϞσϧΛ࡞ΓɺͦΕΒΛൺֱ͢Δ
99Ϛʔέοτ͝ͱʹ༧ଌϞσϧΛ࡞Γ͍ͨͷͰɺʮ܁Γฦ͠ʯʹ ’Market’ ྻΛબͿ
100
• RMSE, MAEͰݟΔͱɺAfrica (ΞϑϦΧ)ͷ΄͏͕Asia Pacific (ΞδΞଠฏ༸Ҭ)ΑΓ༧ଌͱ࣮ଌͷ͕ࠩখ͍͜͞ͱ͕͔Δɻ• ͜ΕΛͬͯɺAfricaͷํ͕Α͘༧ଌͰ͖͍ͯΔͱݴ͑ΔͩΖ͏͔ʁϞσϧͷ༧ଌਫ਼ͷࢦඪΛൺΔ101
• Asia Pacific (ΞδΞଠฏ༸Ҭ) ͷํ͕Africa (ΞϑϦΧ)ΑΓച্ֹ͕େ͖͍ͷͰɺRMSE, MAE͕େ͖͘ͳΔͷͨΓલͱݴ͑Δɻ• ͷεέʔϧʹ͕ࠩ͋Δͱ͖ɺRMSEɺMAEʹΑΔ༧ଌੑೳͷൺֱҙຯΛͳ͞ͳ͍ɻϞσϧͷ༧ଌੑೳͷαϚϦͷදࣔ102
• MAPEɺ༧ଌͱ࣮ଌͷࠩΛɺ࣮ଌͷύʔηϯτͰදͨ͠ͷɻ• ࣮ଌͷͱͱͷେ͖͞ʹؔͳ͘༧ଌੑೳͷൺֱ͕Ͱ͖Δɻ• RMSE, MAEͷେ͖͔ͬͨAsia Pacificͷ΄͏͕ɺMAPEΑΓখ͘͞ɺ࣮AfricaΑΓΑ͍༧ଌ͕Ͱ͖͍ͯͨ͜ͱ͕͔ΔɻϞσϧͷ༧ଌੑೳͷαϚϦͷදࣔ103
• MAPEಉ͡Α͏ʹɺ࣮ଌͷͱͱͷେ͖͞ʹؔͳ͘༧ଌੑೳͷൺֱ͕Ͱ͖Δɻ• MAPEɺ͕0Λ·͍ͨͩΓɺ0ʹۙ͘ͳΔͱ͖ෆ҆ఆʹͳΔ͕ɺMASEʹ͜ͷͳ͍ɻ• MASEͰݟͯɺRMSE, MAEͷେ͖͔ͬͨAsia Pacificͷ΄͏͕ɺ࣮ΑΓΑ͍༧ଌ͕Ͱ͖͍ͯͨ͜ͱ͕͔ΔɻϞσϧͷ༧ଌੑೳͷαϚϦͷදࣔ104
֎෦༧ଌม
ച্ʢSalesʣͱ૬ؔؔʹ͋Δม͕͋ͬͨͱͯ͠ɺ͞ΒʹͦͷมͷকདྷͷΛ༧ଌͰ͖Δɺ·ͨίϯτϩʔϧͰ͖Δͱͨ͠Βɺച্ʢSalesʣΛ͞Βʹྑ͍ਫ਼Ͱ༧ଌͰ͖ΔͷͰͳ͍͔ʁ
Sales ͱ Sales Comp. ૬͍ؔͯ͠ΔΑ͏ͩɻ
Sales ͱ Marketing ૬͍ؔͯ͠ΔΑ͏ͩɻ
Sales ͱ Discount (Avg) ͋·Γ૬ؔͯͦ͠͏ʹͳ͍ɻ
ϚʔέςΟϯά࣍ͷ3ϲ݄ɺͲΕ͚ͩࠂʹ͓ۚΛඅ͔͢ίϯτϩʔϧͰ͖Δͱͨ͠ΒɺͦͷใΛϞσϧʹΈࠐΉ͜ͱͰ͖ͳ͍͔ʁྫ
ఱؾച্ఱؾʹΑͬͯӨڹΛड͚Δͱ͢Δͱɺ࣍ͷ10ؒͷؾԹ߱ਫ֬Λ༧ଌͯ͠ɺͦΕΒΛച্ͷ༧ଌϞσϧʹΈࠐΉͱɺ͞Βʹྑ͍ਫ਼Ͱച্͛Λ༧ଌͰ͖ΔͷͰͳ͍ͩΖ͏͔ɻྫ
มʢྻʣΛ֎෦༧ଌมͱׂͯͯ͠ɺ༧ଌϞσϧΛ࡞Δ͜ͱ͕Ͱ͖ΔɻProphet֎෦༧ଌมʢྫɿϚʔέςΟϯάඅ༻ʣ͕λʔήοτมʢྫɿച্ʣΛ༧ଌ͢ΔͨΊʹཱ͔ͭΛௐɺͦͷ֎෦༧ଌมͷΛ୳͠ग़͢ɻ֎෦༧ଌม
τϨϯυͱपظͷقઅੑΛͱʹͨ͠༧ଌϞσϧ
ϕʔεϞσϧͷධՁ
τϨϯυͱपظͷقઅੑͱച্ใुΛͱʹͨ͠༧ଌϞσϧ
ϞσϧͷධՁϕʔεϞσϧϕʔεϞσϧʹച্ใुΛ͚ͨ͠
τϨϯυͱपظͷقઅੑͱϚʔέςΟϯάඅ༻Λͱʹͨ͠༧ଌϞσϧ
ϕʔεϞσϧʹച্ใुΛ͚ͨ͠ϕʔεϞσϧʹϚʔέςΟϯάඅ༻Λ͚ͨ͠
࣍ͷ3ͭશ෦ͯ͠ΈΔSales Comp., Marketing, and Discount
With Sales Comp., Marketing, Discount
The forecasting model quality has improved for a little bit.ϕʔεϞσϧʹച্ใुΛ͚ͨ͠ϕʔεϞσϧʹച্ใुɺϚʔέςΟϯάඅ༻ɺׂҾΛ͚ͨ͠
ޮՌλϒͷԼͰɺͦΕͧΕͷقઅੑͱ༧ଌม͕༧ଌʹͲͷΑ͏ʹӨڹΛ༩͑Δͷ͔ΛݟΔ͜ͱ͕Ͱ͖Δɻ
มॏཁλϒͷԼͰɺΘΕͨقઅੑͱ༧ଌมͷ͏ͪͲͷม͕ΑΓӨڹ͕ڧ͍ͷ͔ΛݟΔ͜ͱ͕Ͱ͖Δɻ
Q & A
࣍ճηϛφʔ
σʔλαΠΤϯεX3/5 () 1PM (ຊ࣌ؒʣ
5݄։࠵ܾఆʂฏ൛ɿ26ʢՐʣɺ27ʢਫʣɺ28ʢʣि൛ɿ23ʢʣɺ24ʢʣɺ30ʢʣ
• ϓϩάϥϛϯάͳ͠RݴޠͷUIͰ͋ΔExploratoryΛੳπʔϧͱͯ͠༻͢ΔͨΊडߨதɺϏδωεͷΛղܾ͢ΔͨΊʹඞཁͳσʔλαΠΤϯεͷख๏ͷशಘʹ100ˋूதͰ͖Δ• ੳπʔϧͷϕϯμʔϩοΫΠϯͳ͠ExploratoryͰͷ࡞ۀશͯಠཱͨ͠ΦʔϓϯιʔεͷRڥͰ࠶ݱ͕Մೳ• ϏδωεͰ͑ΔࢥߟྗͱεΩϧͷशಘσʔλαΠΤϯεͷεΩϧशಘ͚ͩͰͳ͘ɺσʔλੳʹඞཁͳࢥߟྗशಘͰ͖Δಛ
࿈བྷઌϝʔϧ[email protected]ΣϒαΠτhttps://ja.exploratory.ioϒʔτΩϟϯϓɾτϨʔχϯάhttps://ja.exploratory.io/training-jpTwitter@KanAugust
kanaugust
hygradme
kojira
bengo4com
radkmb
etaroid
akiroom
biwashi
iamshunta
inureyes
kei_s
pospome
tkikuchi1002
bkeepers
gr2m
jensimmons
bcantrill
eileencodes
kevinliebholz
pedronauck
wjessup
deanohume
malarkey
maltzj