Amazon Machine Learning を使ってみた

Transcript

1. Amazon ML Λ
 ࢖ͬͯΈͨ Kenta Murata 2015.04.21

2. ػցֶश

3. ػցֶशͰͰ͖Δ͜ͱ 1. ճؼ 2. ෼ྨ 3. ΫϥελϦϯά

4. ػցֶशͰͰ͖Δ͜ͱ 1. ճؼ 2. ෼ྨ 3. ΫϥελϦϯά → ࣮਺஋ͷ༧ଌ http://commons.wikimedia.org/wiki/File:Linear_regression.svg

   http://commons.wikimedia.org/wiki/File:Polyreg_scheffe.svg

5. ػցֶशͰͰ͖Δ͜ͱ 1. ճؼ 2. ෼ྨ 3. ΫϥελϦϯά → ࣮਺஋ͷ༧ଌ →

   ͔̋×͔Λ༧ଌ http://en.wikipedia.org/wiki/File:SVM_with_soft_margin.pdf

6. ػցֶशͰͰ͖Δ͜ͱ 1. ճؼ 2. ෼ྨ 3. ΫϥελϦϯά → ࣮਺஋ͷ༧ଌ →

   ͔̋×͔Λ༧ଌ → ࣗಈάϧʔϓ෼͚ http://commons.wikimedia.org/wiki/File:KMeans-density-data.svg

7. Amazon Machine Learning

8. Amazon Machine Learning ͰͰ͖Δ͜ͱ 1. ճؼ 2. ೋ஋෼ྨ 3. ଟ஋෼ྨ

9. Amazon Machine Learning ͰͰ͖Δ͜ͱ 1. ճؼ 2. ೋ஋෼ྨ 3. ଟ஋෼ྨ

   ΍ͬͯΈͨ

10. Amazon Machine Learning Ͱ
 ଟ஋෼ྨثΛ࡞Δ

11. σʔλͷ४උ ↓ σʔλιʔε࡞੒ ↓ Ϟσϧ࡞੒ ↓ (σʔλιʔεͷࣗಈ෼ׂ) ↓ Ϟσϧͷֶश ↓

    ϞσϧͷධՁ ଟ஋෼ྨثͷ࡞੒खॱ

12. σʔλͷ४උ

13. None

14. 70,000ݸͷखॻ͖਺ࣈ http://myselph.de/neuralNet.html 28px 28px

15. 60,000ݸ → ֶश༻ 10,000ݸ → ධՁ༻ ֶश༻ͱධՁ༻ʹ༧Ί෼͚ͯ഑෍͞Ε͍ͯΔ

16. όΠφϦσʔλͳͷͰ CSV ΁ม׵͢Δ

17. 28px 28px y, x1, x2,ɾɾɾ, x_k,ɾɾɾ, x784 8, 0, 0,ɾɾɾ,

    221,ɾɾɾ, 0 256֊ௐάϨΠεέʔϧ ਖ਼ղϥϕϧ ϐΫηϧ஋

18. μ΢ϯϩʔυ͢Δ

19. https://rubygems.org/gems/mnist

20. $ gem install mnist $ mnist2csv train-images-idx3-ubyte.gz train-labels-idx1-ubyte.gz > mnist_train.csv

    $ mnist2csv t10k-images-idx3-ubyte.gz t10k-labels-idx1-ubyte.gz > mnist_test.csv

21. CSV ϑΝΠϧΛ S3 ʹΞοϓϩʔυ͢Δ

22. σʔλιʔεΛ࡞Δ

23. None

24. Ξοϓϩʔυͨ͠
 CSV ϑΝΠϧ

25. None
26. None
27. None
28. None

29. ෼ྨର৅ͷΧϥϜΛબ୒ͯ͠Ͷὑ

30. σʔλΛݟͯࣗಈ൑ఆ

31. ༧ଌ݁Ռ͕σʔλιʔεͷͲͷߦʹରԠ͢Δ͔Λ
 ࣝผ͢ΔͨΊͷ ID ͕͋Ε͹ࢦఆ͢Δ ࠓճ͸ແ͍ͷͰࢦఆ͠ͳ͍

32. None
33. None
34. None
35. None

36. ϞσϧΛ࡞Δ

37. None

38. ೖྗσʔλΛબ୒

39. બͿ

40. None
41. None

42. σʔλΛ 7:3 ʹ෼ׂͯ͠ 7 ͷํΛ܇࿅ʹɺ3 ͷํ ΛϞσϧͷධՁʹ࢖͏

43. ͍Ζ͍Ζࣗ෼Ͱࢦఆ͢Δ ࠓճ͸ͬͪ͜

44. None

45. σʔλͷલॲཧํ๏ͳͲ Λ JSON Ͱࢦఆ͢Δ ϑΟʔϧυɻ ࠓճ͸ CSV ʹม׵ͨ͠ ͚ͩͰલॲཧ͕׬ྃͯ͠ ΔͷͰσϑΥϧτͷ··

    Ͱ͓̺
46. None

47. Regularization (ਖ਼ଇԽ) ͸ɺϞσϧͷաֶश (܇࿅σʔ λʹద߹͗ͯ͢͠͠·͏ࣄ) Λ๷͙ͨΊʹߦ͏ɻ L1 (Lasso ճؼ) ͸ɺෆཁͳύϥϝʔλΛ࡟ͬͯϞσϧΛ

    γϯϓϧʹ͍ͨ͠ͱ͖ʹ࢖͏ɻ L2 (Ridge ճؼ) ͸׈Β͔ͳϞσϧ͕ཉ͍͠ͱ͖ʹ࢖͏ɻ (ײ૝: L1 ͱ L2 ΛࠞͥΒΕΕ͹΋ͬͱྑ͍ͷʹ)
48. None

49. Ϟσϧͷ࡞੒ޙʹࣗಈతʹධՁ΋࣮ࢪ͢Δ͔Ͳ͏͔ɻ ࠓճ͸ผʹධՁΛ΍ΔͷͰ No ΛબͿɻ

50. None
51. None

52. ϞσϧΛ࡞Δ

53. ֶशδϣϒ͸ࣗಈతʹ։࢝͢Δ

54. None

55. 60,000 ڭࢣσʔλ → ໿20෼

56. ϞσϧΛධՁ͢Δ

57. None
58. None
59. None
60. None
61. None
62. None
63. None

64. 10,000 ςετσʔλ → 1ʙ2෼

65. None

66. ҎԼͷࣜͰܭࢉ͞ΕΔϞσϧͷ༏ल͞ΛଌΔྔ 2 × ద߹཰ × ࠶ݱ཰
 ద߹཰ + ࠶ݱ཰

67. ਅͷ෼ྨ 1 ͦͷଞ ༧
 ଌ
 ݁
 Ռ 1 True Positive

    False Positive ͦ
 ͷ
 ଞ False Negative True Negative ద߹཰ ʹ ࠶ݱ཰ ʹ True Positive
 True Positive + False Positive True Positive
 True Positive + False Negative TP FP FN TN TP FP FN TN
68. None

69. 1,000 ڭࢣσʔλͰ࡞ͬͨϞσϧͷ৔߹

70. None

71. ڭࢣσʔλ͕ଟ͍΄ͲϞσϧͷੑೳ͕ྑ͘ͳΔ

72. ϞσϧΛ࢖͏

73. Ϟσϧͷ࢖͍ํ 1. όον༧ଌ 2. ϦΞϧλΠϜ༧ଌ

74. Ϟσϧͷ࢖͍ํ 1. όον༧ଌ 2. ϦΞϧλΠϜ༧ଌ → ·ͱ·ͬͨσʔλΛ·ͱΊͯ༧ଌ

75. Ϟσϧͷ࢖͍ํ 1. όον༧ଌ 2. ϦΞϧλΠϜ༧ଌ → ·ͱ·ͬͨσʔλΛ·ͱΊͯ༧ଌ → API Λ࢖ͬͯ1ͭͣͭ༧ଌ

76. Amazon Machine Learning ͷྉۚମܥ

77. Amazon Machine Learning ͷྉۚମܥ

78. 1,000 σʔλͰϞσϧΛ࡞ͬͨͱ͖

79. 70,000 σʔλͰϞσϧΛ࡞ͬͨͱ͖

80. S3 price

81. Amazon Machine Learning Λ࢖ͬͯΈͨײ૝ 1. Α͘Ͱ͖ͯΔ 2. ͬ͘͞ͱϓϩτλΠϓ͍ͨ࣌͠ʹศརͦ͏ 3. ֶशࡁΈͷϞσϧΛΤΫεϙʔτͰ͖ͳ͍

82. Amazon Machine Learning Λ࢖ͬͯΈͨײ૝ 1. Α͘Ͱ͖ͯΔ 2. ͬ͘͞ͱϓϩτλΠϓ͍ͨ࣌͠ʹศརͦ͏ → ΞϧΰϦζϜΛදʹग़ͣ͞ʹ্ख͘؆ུԽͯ͠Δ

    3. ֶशࡁΈͷϞσϧΛΤΫεϙʔτͰ͖ͳ͍

83. Amazon Machine Learning Λ࢖ͬͯΈͨײ૝ 1. Α͘Ͱ͖ͯΔ 2. ͬ͘͞ͱϓϩτλΠϓ͍ͨ࣌͠ʹศརͦ͏ → ΞϧΰϦζϜΛදʹग़ͣ͞ʹ্ख͘؆ུԽͯ͠Δ

    → ࣮ӡ༻લʹ༷ʑͳಛ௃ϕΫτϧΛ؆୯ʹࢼͤΔ 3. ֶशࡁΈͷϞσϧΛΤΫεϙʔτͰ͖ͳ͍

84. Amazon Machine Learning Λ࢖ͬͯΈͨײ૝ 1. Α͘Ͱ͖ͯΔ 2. ͬ͘͞ͱϓϩτλΠϓ͍ͨ࣌͠ʹศརͦ͏ → ΞϧΰϦζϜΛදʹग़ͣ͞ʹ্ख͘؆ུԽͯ͠Δ

    → ࣮ӡ༻લʹ༷ʑͳಛ௃ϕΫτϧΛ؆୯ʹࢼͤΔ 3. ֶशࡁΈͷϞσϧΛΤΫεϙʔτͰ͖ͳ͍ → ࣮ӡ༻࣌͸ࣗ෼Ͱ࣮૷ͨ͠ϞσϧΛ࢖͏
 ɹ ϓϩτλΠϓͰ্ख͘ߦ͖ͦ͏ͳ͜ͱ͕
 ɹ ෼͔ͬͯΔͷͰ࣮૷ίετ΋ؾʹͳΒͳ͍!?