Exploratory: 「距離」とMDS(多次元尺度構成法)のアルゴリズムを使って類似性を可視化する

Exploratory: 「距離」とMDS(多次元尺度構成法)のアルゴリズムを使って類似性を可視化する

カスタマー、国、製品といった、興味のある対象間にある類似性を理解するための「距離」のアルゴリズムと、そういった関係性を直感的に理解するために可視化するための「多次元尺度構成法(Multi-Dimensional Scaling / MDS) 」というアルゴリズムの紹介と、それらのExploratoryでの使い方の話をします。

19fc8f6113c5c3d86e6176362ff29479?s=128

Kan Nishida

July 17, 2019
Tweet

Transcript

  1. 6.

    ୈ1ͷ೾ ୈ̎ͷ೾ ୈ̏ͷ೾ ϓϥΠϕʔτ(ߴ͍/ݹ͍) Φʔϓϯɾιʔε(ແྉ/࠷ઌ୺) UI & ϓϩάϥϛϯά ϓϩάϥϛϯά 2016

    2000 1976 ϚωλΠθʔγϣϯ ίϞσΟςΟԽ ຽओԽ ౷ܭֶऀ σʔλαΠΤϯςΟετ Exploratory ΞϧΰϦζϜ Ϣʔβʔɾ ମݧ πʔϧ Φʔϓϯɾιʔε(ແྉ/࠷ઌ୺) UI & ࣗಈԽ ϏδωεɾϢʔβʔ ςʔϚ σʔλαΠΤϯεͷຽओԽ
  2. 11.

    11

  3. 14.

    14 • ΤϧαϨϜ: ૯ձܾٞ ES-10/L.22 - ΞϝϦΧͷΤϧαϨϜ ʹؔ͢Δ੓ࡦ΁ͷඇ೉ (2017) •

    ΢ΫϥΠφ: ૯ձܾٞ 68/262 - ΢ΫϥΠφͷྖ౔อશ (2014) ྫ) 2ͭͷࠃ࿈ܾٞ
  4. 15.

    15 ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ No Yes ϩγΞ Yes No Χφμ

    غݖ Yes ֤ࠃͷ͜ΕΒͷܾٞ΁ͷ౤ථ݁Ռ
  5. 16.

    16 • Yes -> 1 • No -> -1 •

    غݖ -> 0 ౤ථΛ਺஋Խ͢Δ
  6. 17.

    17 ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ -1 1 ϩγΞ 1 -1 Χφμ

    0 1 ֤ࠃͷ͜ΕΒͷܾٞ΁ͷ౤ථ݁Ռ
  7. 18.

    18 ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ ϩγΞ ( -1, 1 ) (

    1, -1 ) 1 (Yes) -1 (No) 0 -1 (No) 1 (Yes) ౤ථ݁ՌΛάϥϑ্ʹϓϩοτ Χφμ
  8. 19.

    19 ( -1, 1 ) ( 1, -1 ) 2.828

    1 (Yes) -1 (No) 0 -1 (No) 1 (Yes) ΞϝϦΧͱϩγΞͷؒͷϢʔΫϦουڑ཭ ௚ઢڑ཭ ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ ϩγΞ Χφμ
  9. 20.

    20 ( -1, 1 ) ( 1, -1 ) 1

    1 (Yes) 1 (Yes) -1 (No) 0 -1 (No) ( 0 , 1 ) ΞϝϦΧͱΧφμͷؒͷϢʔΫϦουڑ཭ ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ ϩγΞ Χφμ
  10. 21.

    21 ( -1, 1 ) ( 1, -1 ) 1

    1 (Yes) 1 (Yes) -1 (No) 0 -1 (No) ( 0 , 1 ) ϢʔΫϦουڑ཭ 2.828 ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ ϩγΞ Χφμ
  11. 23.

    23 ( -1, 1 ) ( 1, -1 ) 1

    (Yes) -1 (No) 0 -1 (No) 4 1 (Yes) ΞϝϦΧͱϩγΞͷؒͷϚϯϋολϯڑ཭ άϦουͰڑ཭Λ΋ͱΊΔ ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ ϩγΞ Χφμ
  12. 24.

    24 ( -1, 1 ) ( 1, -1 ) 1

    1 (Yes) -1 (No) 0 -1 (No) ( 0 , 1 ) 1 (Yes) ΞϝϦΧͱΧφμͷؒͷϚϯϋολϯڑ཭ ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ ϩγΞ Χφμ
  13. 25.

    25 ϢʔΫϦουڑ཭ ( -1, 1 ) ( 1, -1 )

    ( -1, 1 ) ( 1, -1 ) ( 0 , 1 ) ( 0 , 1 ) 4 2.828 1 1 Ϛϯϋολϯڑ཭ 1 -1 -1 0 1 -1 -1 0 1 1 ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ ϩγΞ Χφμ ΤϧαϨϜ ΢ΫϥΠφ ΞϝϦΧ ϩγΞ Χφμ
  14. 30.

    30 1 RC_ID ΞϝϦΧ ೔ຊ 1 1 1 2 1

    0 3 0 1 4 0 0 … δϟΧʔυ܎਺ ೔ຊ΋ΞϝϦΧ΋1 ΞϝϦΧ͚ͩ1 ೔ຊ͚ͩ1 5 2 6 4 3 7
  15. 31.

    31 1 RC_ID ΞϝϦΧ Χφμ 1 1 1 2 1

    1 3 1 1 4 1 0 … ڑ཭͕͍ۙ৔߹ Χφμ΋ΞϝϦΧ΋1 ΞϝϦΧ͚ͩ1 Χφμ͚ͩ1 5 2 6 4 3 7
  16. 32.

    32 1 RC_ID ΞϝϦΧ ϩγΞ 1 1 0 2 1

    0 3 0 1 4 1 0 … ڑ཭͕ԕ͍৔߹ ϩγΞ΋ΞϝϦΧ΋1 ΞϝϦΧ͚ͩ1 ϩγΞ͚ͩ1 5 2 6 4 3 7
  17. 33.

    33 δϟΧʔυ܎਺ όΠφϦڑ཭ = 1 - δϟΧʔυ܎਺ όΠφϦڑ཭ 0~1 ࣅͯΔ΄Ͳ̍ʹۙͮ͘

    ҟͳ͍ͬͯΔ΄Ͳ̍ʹۙͮ͘ 0~1 ࣅ͍ͯΔ΄Ͳ̌ʹۙͮ͘ ҟͳ͍ͬͯΔ΄Ͳ̍ʹۙͮ͘
  18. 38.
  19. 43.

    43 େࡕ ౦ژ 500km େࡕ ژ౎ 60km 460km ౦ژ ژ౎

    ௚ઢ্ʹՄࢹԽ͢ΔͱɺଟগΘ͔Γ΍͘͢ͳΔ
  20. 54.
  21. 56.

    Month NY CA FL TX WA Jan 15 5 10

    20 10 Feb 50 30 25 40 24 ϫΠυܕ
  22. 57.

    Month NY CA FL TX WA OR MT MI NJ

    NV Jan 15 5 10 20 10 10 40 79 20 60 Feb 50 30 25 40 24 6 15 55 9 5 Mar 10 3 14 0 4 5 20 5 2 4 Apr 3 79 20 60 5 10 20 5 0 20 May 55 55 9 5 30 25 40 5 10 20 Y1 X Y2 Y3 Yn X: Month NY Y1: CA Y2: Jan Feb Mar Apr May Jun
  23. 58.

    Month NY CA FL TX WA Jan 15 5 10

    20 10 Feb 50 30 25 40 24 ϫΠυܕ Month State NumBabies Jan NY 15 Feb NY 50 Jan CA 5 Feb CA 30 Jan FL 10 Feb FL 25 Jan TX 20 Feb TX 40 Jan WA 10 Feb WA 24 ϩϯάܕ
  24. 59.

    Month NY CA FL TX WA Jan 15 5 10

    20 10 Feb 50 30 25 40 24 ϫΠυܕ
  25. 60.

    Month NY CA FL TX WA Jan 15 5 10

    20 10 Feb 50 30 25 40 24 ϫΠυܕ Month State NumBabies Jan NY 15 Feb NY 50 Jan CA 5 Feb CA 30 Jan FL 10 Feb FL 25 Jan TX 20 Feb TX 40 Jan WA 10 Feb WA 24 ϩϯάܕ Gather
  26. 61.

    Month NY CA FL TX WA Jan 15 5 10

    20 10 Feb 50 30 25 40 24 Month State NumBabies Jan NY 15 Feb NY 50 Jan CA 5 Feb CA 30 Jan FL 10 Feb FL 25 gather(State, NumBabies, NY:WA)
  27. 69.

    Month State NumBabies Jan NY 15 Feb NY 50 Jan

    CA 5 Feb CA 30 Jan FL 10 Feb FL 25 Jan TX 20 Feb TX 40 Jan WA 10 Feb WA 24 ؍ଌ(Observation) ม਺(Variables) Tidyσʔλ
  28. 70.

    Month NY CA FL TX WA Jan 15 5 10

    20 10 Feb 50 30 25 40 24 ϫΠυܕ Month State NumBabies Jan NY 15 Feb NY 50 Jan CA 5 Feb CA 30 Jan FL 10 Feb FL 25 Jan TX 20 Feb TX 40 Jan WA 10 Feb WA 24 ϩϯάܕ Spread, Pivot, Un-Tidy
  29. 71.

    Month NY CA FL TX WA Jan 15 5 10

    20 10 Feb 50 30 25 40 24 Month State NumBabies Jan NY 15 Feb NY 50 Jan CA 5 Feb CA 30 Jan FL 10 Feb FL 25 spread(State, NumBabies)
  30. 76.
  31. 77.

    77 ࣌ؒ λΠτϧ ൃදऀ 19:00 Exploratory v5.3ͷ঺հ ੢ాצҰ࿠ 19:30 ౷ܭϞσϧΛར༻ͨ͠ैۀһͷຬ଍౓෼ੳ

    େฏ༟ี 19:50 ExploratoryΛ׆༻ͨࣾ͠಺σʔλαΠΤϯεษڧձͷऔΓ૊Έ ໦ݪ༞հ 20:10 ػցֶशΛ༻͍ͨϚʔέςΟϯάߪങཁҼ෼ੳ सࢁ༟ଠ 20:30 EDA Salonͷ঺հ ଜཬҮ࠸ 21:00 ࠙਌ձ ΞδΣϯμ
  32. 79.
  33. 83.
  34. 85.