iOSエンジニアのための、SwiftからPythonのライブラリを使って機械学習する方法 / Machine Learning using Python from Swift for iOS Engineer

Swift 4.2では他言語のAPIをSwiftから簡単に呼び出せるようになります。これ自体は汎用的な仕組みですが、最大の目的はSwiftからPythonの資産を活用できるようにし、Python並の機械学習を実現することのようです。日々Swiftを書いているiOSエンジニアにとってこれは大きなチャンスです。このトークではiOSエンジニアに向けて、SwiftのコードでPythonのライブラリを使って機械学習する方法を紹介します。

Transcript

1. iOS ΤϯδχΞͷͨΊͷɺ
   Swi$ ͔Β Python ͷϥΠϒϥϦΛ
   ࢖ͬͯػցֶश͢Δํ๏
   Yuta Koshizawa @koher
   

   View Slide

2. View Slide

3. !
   ػցֶशೖ໳
   

   View Slide

4. View Slide

5. !
   ෼ྨʢ Classifica)on ʣ
   

   View Slide

6. ྫ: ݂ӷݕࠪʹΑΔഏ͕Μͷ਍அ
   

   View Slide

7. ྫ: ݂ӷݕࠪʹΑΔഏ͕Μͷ਍அ
   ݈߁ͳਓͷσʔλ
   ඃݕऀ ID CYFRA (ng/ml) CEA (ng/ml)
   0001 3.0 4.0
   0002 2.8 4.5
   0003 3.1 3.7
   ... ... ...
   

   View Slide

8. ྫ: ݂ӷݕࠪʹΑΔഏ͕Μͷ਍அ
   ഏ͕Μͷਓͷσʔλ
   ඃݕऀ ID CYFRA (ng/ml) CEA (ng/ml)
   1001 4.0 6.0
   1002 3.4 7.0
   1003 4.9 4.8
   ... ... ...
   

   View Slide

9. ID: 0001, CYFRA: 3.0 ng/ml, CEA: 4.0 ng/ml
   

   View Slide

10. ID: 0002, CYFRA: 2.8 ng/ml, CEA: 4.5 ng/ml
    

    View Slide

11. ID: 0003, CYFRA: 3.1 ng/ml, CEA: 3.7 ng/ml
    

    View Slide

12. ɾɾɾ
    

    View Slide

13. ID: 1001, CYFRA: 4.0 ng/ml, CEA: 6.0 ng/ml
    

    View Slide

14. ID: 1002, CYFRA: 3.4 ng/ml, CEA: 7.0 ng/ml
    

    View Slide

15. ID: 1003, CYFRA: 4.9 ng/ml, CEA: 4.8 ng/ml
    

    View Slide

16. ɾɾɾ
    

    View Slide

17. ઢͷԼ → ݈߁ɺઢͷ্ → ഏ͕Μ
    

    View Slide

18. CYFRA: 3.2 ng/ml, CEA: 3.8 ng/ml
    

    View Slide

19. CYFRA: 4.5 ng/ml, CEA: 5.9 ng/ml
    

    View Slide

20. ػցֶश
    !
    → ྑͦ͞͏ͳ௚ઢΛܾΊΔ
    

    View Slide

21. y = a x + b
    

    View Slide

22. y = a x + b
    

    View Slide

23. y = a x + b
    

    View Slide

24. y = -1.78 x + 11.1
    

    View Slide

25. ௚ઢͰ෼ྨ
    

    View Slide

26. ௚ઢͰ෼ྨ
    

    View Slide

27. ͜Μͳσʔλͩͱʁ
    

    View Slide

28. ௚ઢͰ෼ྨ͢Δͷ͸ແཧ͋Γ
    

    View Slide

29. ͜ΜͳۂઢͰ෼ྨ͍ͨ͠
    

    View Slide

30. ௚ઢͷࣜͰ͸ແཧ
    

    View Slide

31. ΑΓෳࡶͳ͕ࣜඞཁ
    

    View Slide

32. ෳࡶͳํ͕දݱྗ͕ߴͯ͘ྑ͍ʁ
    

    View Slide

33. ͞ΒʹෳࡶͳࣜʹΑΔۂઢ
    

    View Slide

34. CYFRA: 3.2 ng/ml, CEA: 4.7 ng/ml
    

    View Slide

35. CYFRA: 3.2 ng/ml, CEA: 4.7 ng/ml
    

    View Slide

36. !
    ෳࡶͳϞσϧͳΒྑ͍ͱ͍͏Θ͚Ͱ͸ͳ͍
    

    View Slide

37. !
    ൚Խʢ Generaliza*on ʣ
    

    View Slide

38. ͲͷΑ͏ʹ൚ԽೳྗΛධՁ͢Ε͹Α͍ʁ
    

    View Slide

39. !
    !
    ܇࿅σʔλʢ Training data ʣ
    !
    ςετσʔλʢ Test data ʣ
    ʹ෼͚ͯ൚ԽೳྗΛධՁ͢Δɻ
    

    View Slide

40. શσʔλ
    

    View Slide

41. ܇࿅σʔλ
    

    View Slide

42. ܇࿅σʔλΛ࢖ֶͬͯश
    

    View Slide

43. ܇࿅σʔλ: ਫ਼౓ 100%
    

    View Slide

44. ςετσʔλΛ࢖ͬͯධՁ
    

    View Slide

45. ςετσʔλ: ਫ਼౓ 90.0%
    

    View Slide

46. ܇࿅σʔλΛ࢖ֶͬͯश
    

    View Slide

47. ܇࿅σʔλ: ਫ਼౓ 97.1%
    

    View Slide

48. ςετσʔλΛ࢖ͬͯධՁ
    

    View Slide

49. ςετσʔλ: ਫ਼౓ 95.8%
    

    View Slide

50. ܇࿅σʔλΛ࢖ֶͬͯश
    

    View Slide

51. ܇࿅σʔλ: ਫ਼౓ 89.6%
    

    View Slide

52. ςετσʔλΛ࢖ͬͯධՁ
    

    View Slide

53. ςετσʔλ: ਫ਼౓ 85.8%
    

    View Slide

54. View Slide

55. View Slide

56. ܇࿅σʔλΛ 10 ഒʹ૿΍ͯ͠ΈΔ
    

    View Slide

57. Overfi&ng ͷͱ͖ʢ܇࿅σʔλʣ
    

    View Slide

58. Overfi&ng ͷͱ͖ʢ܇࿅σʔλʣ
    

    View Slide

59. Overfi&ng ͷͱ͖ʢςετσʔλ: 90.0%ʣ
    

    View Slide

60. Overfi&ng ͷͱ͖ʢςετσʔλ: 95.0%ʣ
    

    View Slide

61. Underfi'ng ͷͱ͖ʢ܇࿅σʔλʣ
    

    View Slide

62. Underfi'ng ͷͱ͖ʢ܇࿅σʔλʣ
    

    View Slide

63. Underfi'ng ͷͱ͖ʢςετσʔλ: 85.8%ʣ
    

    View Slide

64. Underfi'ng ͷͱ͖ʢςετσʔλ: 85.8%ʣ
    

    View Slide

65. !
    !
    ద߹ෆ଍ʢ Underfi(ng ʣ
    !
    ա৒ద߹ʢ Overfi'ng ʣ
    Λ۠ผͯ͠ରԠํ๏Λߟ͑Δ
    

    View Slide

66. ಛ௃ྔ͕ 3 ݸ͋Δ৔߹
    ඃݕऀ ID CYFRA (ng/ml) CEA (ng/ml) SCC (ng/ml)
    0001 3.0 4.0 1.1
    0002 2.8 4.5 1.4
    0003 3.1 3.7 0.9
    ... ... ... ...
    

    View Slide

67. ಛ௃ྔ͕ 3 ݸ͋Δ৔߹
    

    View Slide

68. ಛ௃ྔ͕ 3 ݸ͋Δ৔߹
    

    View Slide

69. !
    4 ࣍ݩɾߴ࣍ݩۭؒΛ
    Πϝʔδ͢Δํ๏
    

    View Slide

70. ࣍ݩ
    

    View Slide

71. ෼ྨͷͨΊʹ͸ۭؒΛ෼ׂ͢Δ
    

    View Slide

72. ࣍ݩͱஅ໘
    

    View Slide

73. ࣍ݩͱஅ໘
    

    View Slide

74. ࣍ݩͱஅ໘
    

    View Slide

75. ࣍ݩͱஅ໘
    

    View Slide

76. !
    ௒ۂ໘ʢ Hypersurface ʣ
    

    View Slide

77. ۂ͕ͬͯͳ͍ͱ͖͸
    !
    ௒ฏ໘ʢ Hyperplane ʣ
    

    View Slide

78. !
    ๏ଇੑΛݟ͚ͭग़͢
    N ࣍ݩۭؒ͸ N - 1 ࣍ݩۭؒͰ෼ׂͰ͖Δ
    

    View Slide

79. ۭؒʢ 3 ࣍ݩʣ + ࣌ؒʢ 1 ࣍ݩʣ = 4 ࣍ݩʁ
    

    View Slide

80. 4 ࣍ݩۭ࣌ؒͷஅ໘
    

    View Slide

81. ݱࡏ͸
    4 ࣍ݩۭ࣌ؒΛ
    աڈͱະདྷʹ෼͚Δ௒ฏ໘
    

    View Slide

82. !
    4 ࣍ݩ͸ۭ࣌ؒͰߟ͑Δͱ૝૾͠΍͍͢
    

    View Slide

83. 2 ࣍ݩۭؒΛۂઢͰ෼͚Δ͜ͱΛߟ͑Δ
    

    View Slide

84. 10 × 10 ͷ֨ࢠঢ়ͷσʔλ
    

    View Slide

85. 10 × 10 ͷ֨ࢠঢ়ͷσʔλ
    

    View Slide

86. ֨ࢠঢ়ͷσʔλͱ࣍ݩ
    • 2 ࣍ݩͩͱ ݅
    • 3 ࣍ݩͩͱ ݅
    • 4 ࣍ݩͩͱ ݅
    • ...
    • 80 ࣍ݩͩͱ ݅
    

    View Slide

87. ֨ࢠঢ়ͷσʔλͱ࣍ݩ
    • 2 ࣍ݩͩͱ ݅
    • 3 ࣍ݩͩͱ ݅
    • 4 ࣍ݩͩͱ ݅
    • ...
    • 80 ࣍ݩͩͱ ݅ ← ؍ଌՄೳͳӉ஦ʹ͋Δݪࢠͷ਺
    

    View Slide

88. !
    ߴ࣍ݩۭؒ͸ലେ
    ୯७ʹσʔλ݅਺Λ૿΍͢ํ๏Ͱ௅Ήͷ͸ແ๳
    

    View Slide

89. खॻ͖਺ࣈͷ෼ྨ 1
    1 h$p:/
    /scikit-learn.org/stable/datasets/index.html#op;cal-recogni;on-of-handwri$en-digits-data-set
    

    View Slide

90. 8 × 8 = 64 ըૉ
    

    View Slide

91. ೖྗ͸ 64 ࣍ݩͷσʔλ
    

    View Slide

92. ߴ࣍ݩۭؒͰͷ෼ྨͷΠϝʔδ
    

    View Slide

93. ߴ࣍ݩۭؒͰͷ෼ྨͷΠϝʔδ
    

    View Slide

94. ߴ࣍ݩۭؒͰͷ෼ྨͷΠϝʔδ
    

    View Slide

95. ߴ࣍ݩۭؒͰͷ෼ྨͷΠϝʔδ
    

    View Slide

96. ߴ࣍ݩۭؒͰͷ෼ྨͷΠϝʔδ
    

    View Slide

97. !
    ߴ࣍ݩͰ͸ઢܗϞσϧ΋༗༻
    

    View Slide

98. खॻ͖਺ࣈͷ෼ྨʢ Python ʣ
    # scikit-learn ͷؔ਺΍ΫϥεΛ import
    from sklearn.datasets import load_digits
    from sklearn.svm import LinearSVC
    from sklearn.model_selection import train_test_split
    # ֶश
    classifier = LinearSVC()
    dataset = load_digits()
    X_train, X_test, y_train, y_test = \
    train_test_split(dataset["data"], dataset["target"])
    classifier.fit(X_train, y_train)
    # ධՁ
    print(f"train: {classifier.score(X_train, y_train)}")
    print(f"test: {classifier.score(X_test, y_test)}")
    

    View Slide

99. खॻ͖਺ࣈͷ෼ྨʢ Python ʣ
    # scikit-learn ͷؔ਺΍ΫϥεΛ import
    from sklearn.datasets import load_digits
    from sklearn.svm import LinearSVC
    from sklearn.model_selection import train_test_split
    # ֶश
    classifier = LinearSVC()
    dataset = load_digits()
    X_train, X_test, y_train, y_test = \
    train_test_split(dataset["data"], dataset["target"])
    classifier.fit(X_train, y_train)
    # ධՁ
    print(f"train: {classifier.score(X_train, y_train)}")
    print(f"test: {classifier.score(X_test, y_test)}")
    

    View Slide

100. खॻ͖਺ࣈͷ෼ྨʢ Python ʣ
     # scikit-learn ͷؔ਺΍ΫϥεΛ import
     from sklearn.datasets import load_digits
     from sklearn.svm import LinearSVC
     from sklearn.model_selection import train_test_split
     # ֶश
     classifier = LinearSVC()
     dataset = load_digits()
     X_train, X_test, y_train, y_test = \
     train_test_split(dataset["data"], dataset["target"])
     classifier.fit(X_train, y_train)
     # ධՁ
     print(f"train: {classifier.score(X_train, y_train)}")
     print(f"test: {classifier.score(X_test, y_test)}")
     

     View Slide

101. खॻ͖਺ࣈͷ෼ྨʢ Python ʣ
     # scikit-learn ͷؔ਺΍ΫϥεΛ import
     from sklearn.datasets import load_digits
     from sklearn.svm import LinearSVC
     from sklearn.model_selection import train_test_split
     # ֶश
     classifier = LinearSVC()
     dataset = load_digits()
     X_train, X_test, y_train, y_test = \
     train_test_split(dataset["data"], dataset["target"])
     classifier.fit(X_train, y_train)
     # ධՁ
     print(f"train: {classifier.score(X_train, y_train)}")
     print(f"test: {classifier.score(X_test, y_test)}")
     

     View Slide

102. खॻ͖਺ࣈͷ෼ྨʢ Python ʣ
     # scikit-learn ͷؔ਺΍ΫϥεΛ import
     from sklearn.datasets import load_digits
     from sklearn.svm import LinearSVC
     from sklearn.model_selection import train_test_split
     # ֶश
     classifier = LinearSVC()
     dataset = load_digits()
     X_train, X_test, y_train, y_test = \
     train_test_split(dataset["data"], dataset["target"])
     classifier.fit(X_train, y_train)
     # ධՁ
     print(f"train: {classifier.score(X_train, y_train)}")
     print(f"test: {classifier.score(X_test, y_test)}")
     

     View Slide

103. खॻ͖਺ࣈͷ෼ྨʢ Python ʣ
     # scikit-learn ͷؔ਺΍ΫϥεΛ import
     from sklearn.datasets import load_digits
     from sklearn.svm import LinearSVC
     from sklearn.model_selection import train_test_split
     # ֶश
     classifier = LinearSVC()
     dataset = load_digits()
     X_train, X_test, y_train, y_test = \
     train_test_split(dataset["data"], dataset["target"])
     classifier.fit(X_train, y_train)
     # ධՁ
     print(f"train: {classifier.score(X_train, y_train)}")
     print(f"test: {classifier.score(X_test, y_test)}")
     

     View Slide

104. खॻ͖਺ࣈͷ෼ྨʢ݁Ռʣ
     train: 0.9925760950259837
     test: 0.9666666666666667
     

     View Slide

105. !
     Swi% ͔Βಈతܕ෇ݴޠʹ
     ࿈ܞ͢Δͷʹ͔ܽͤͳ͍৽ػೳ
     

     View Slide

106. SE-0195: @dynamicMemberLookup 2
     2 h$ps:/
     /github.com/apple/swi6-evolu9on/blob/master/proposals/0195-dynamic-member-lookup.md
     

     View Slide

107. SE-0216: @dynamicCallable 3
     3 h$ps:/
     /github.com/apple/swi6-evolu9on/blob/master/proposals/0216-dynamic-callable.md
     

     View Slide

108. ੩తܕ෇ݴޠͱͷ࿈ܞ
     import UIKit
     class ViewController: UIViewController {
     override func viewDidLoad() {
     super.viewDidLoad()
     }
     }
     

     View Slide

109. ੩తܕ෇ݴޠͱͷ࿈ܞ
     import UIKit
     class ViewController: UIViewController {
     override func viewDidLoad() {
     super.viewDidLoab() // ίϯύΠϧΤϥʔ
     // ^
     }
     }
     

     View Slide

110. ಈతܕ෇ݴޠͱͷ࿈ܞ
     import NumPy
     let a = array([[1, 2], [3, 4]])
     print(a.shape)
     

     View Slide

111. ಈతܕ෇ݴޠͱͷ࿈ܞ
     import NumPy
     let a = array([[1, 2], [3, 4]])
     print(a.shabe) // ͜Ε͕ΤϥʔʹͳΔ͔࣮ߦ࣌·ͰΘ͔Βͳ͍
     // ^
     

     View Slide

112. ಈతܕ෇ݴޠͱͷ࿈ܞ
     import NumPy
     let a = array([[1, 2], [3, 4]])
     print(a["shape"]) // ͜Μͳײ͕͡ Swift ʹೃછΈͦ͏
     

     View Slide

113. ಈతܕ෇ݴޠͱͷ࿈ܞ
     import NumPy
     let a = array([[1, 2], [3, 4]])
     print(a.shape) // Ͱ΋͜͏ॻ͚ͳ͍ͱਏ͍
     

     View Slide

114. SE-0195: @dynamicMemberLookup
     _ = foo.bar
     foo.bar = x
     // @dynamicMemberLookup Ͱ↑͕↓ͷγϯλοΫεγϡΨʔʹͳΔ
     _ = foo[dynamicMember: "bar"]
     foo[dynamicMember: "bar"] = x
     

     View Slide

115. SE-0195: @dynamicMemberLookup
     @dynamicMemberLookup
     struct Foo {
     subscript(dynamicMember member: String) -> Foo {
     ...
     }
     }
     

     View Slide

116. SE-0216: @dynamicCallable
     foo(42, "XYZ")
     foo(a: 42, b: "XYZ")
     // @dynamicCallable Ͱ↑͕↓ͷγϯλοΫεγϡΨʔʹͳΔ
     foo.dynamicallyCall(withArguments: [42, "XYZ"])
     foo.dynamicallyCall(withKeywordArguments: ["a": 42, "b": "XYZ"])
     

     View Slide

117. SE-0216: @dynamicCallable
     @dynamicCallable
     struct Foo {
     ...
     @discardableResult
     func dynamicallyCall(withArguments: [FooConvertible]) -> Foo {
     ...
     }
     @discardableResult
     func dynamicallyCall(withKeywordArguments: [String: FooConvertible]) -> Foo {
     ...
     }
     }
     

     View Slide

118. Swi$ for TensorFlow 4
     4 h$ps:/
     /medium.com/tensorflow/introducing-swi8-for-tensorflow-b75722c58df0
     

     View Slide

119. !
     Swi% ͔Β Python ͷ
     ϥΠϒϥϦΛ࢖͏ํ๏
     

     View Slide

120. Python
     import numpy as np
     a = np.array([[1, 2, 3], [4, 5, 6]])
     print(a.shape)
     

     View Slide

121. Swi$
     import Python
     let np = Python.import("numpy")
     let a = np.array([[1, 2, 3], [4, 5, 6]])
     print(a.shape)
     

     View Slide

122. View Slide

123. Swi$
     import Python
     let np = Python.import("numpy")
     let a = np.array([[1, 2, 3], [4, 5, 6]])
     print(a.shape)
     

     View Slide

124. Swi$
     import Python
     let np: PythonObject = Python.import("numpy")
     let a: PythonObject = np.array([[1, 2, 3], [4, 5, 6]])
     print(a.shape)
     

     View Slide

125. PythonObject 5
     @dynamicCallable
     @dynamicMemberLookup
     public struct PythonObject {
     ...
     }
     5 h$ps:/
     /github.com/apple/swi6/blob/tensorflow/stdlib/public/Python/Python.swi6
     

     View Slide

126. PythonObject 5
     public extension PythonObject {
     subscript(dynamicMember memberName: String) -> PythonObject {
     get {
     guard let member = checking[dynamicMember: memberName] else {
     fatalError("Could not access PythonObject member '\(memberName)'")
     }
     return member
     }
     nonmutating set {
     let selfObject = ownedPyObject
     defer { Py_DecRef(selfObject) }
     let valueObject = newValue.ownedPyObject
     defer { Py_DecRef(valueObject) }
     if PyObject_SetAttrString(selfObject, memberName, valueObject) == -1 {
     try! throwPythonErrorIfPresent()
     fatalError(...)
     }
     }
     }
     ...
     }
     5 h$ps:/
     /github.com/apple/swi6/blob/tensorflow/stdlib/public/Python/Python.swi6
     

     View Slide

127. PythonObject 5
     public extension PythonObject {
     subscript(dynamicMember memberName: String) -> PythonObject {
     get {
     guard let member = checking[dynamicMember: memberName] else {
     fatalError("Could not access PythonObject member '\(memberName)'")
     }
     return member
     }
     nonmutating set {
     let selfObject = ownedPyObject
     defer { Py_DecRef(selfObject) }
     let valueObject = newValue.ownedPyObject
     defer { Py_DecRef(valueObject) }
     if PyObject_SetAttrString(selfObject, memberName, valueObject) == -1 {
     try! throwPythonErrorIfPresent()
     fatalError(...)
     }
     }
     }
     ...
     }
     5 h$ps:/
     /github.com/apple/swi6/blob/tensorflow/stdlib/public/Python/Python.swi6
     

     View Slide

128. !
     Swi% ͔Β Python ͷ
     ϥΠϒϥϦΛ࢖ͬͯػցֶश
     

     View Slide

129. खॻ͖਺ࣈͷ෼ྨʢ Python ʣ
     # Python
     # scikit-learn ͷؔ਺΍ΫϥεΛ import
     from sklearn.datasets import load_digits
     from sklearn.svm import LinearSVC
     from sklearn.model_selection import train_test_split
     # ֶश
     classifier = LinearSVC()
     dataset = load_digits()
     X_train, X_test, y_train, y_test = \
     train_test_split(dataset["data"], dataset["target"])
     classifier.fit(X_train, y_train)
     # ධՁ
     print(f"train: {classifier.score(X_train, y_train)}")
     print(f"test: {classifier.score(X_test, y_test)}")
     

     View Slide

130. खॻ͖਺ࣈͷ෼ྨʢ Swi% ʣ
     import Python
     // scikit-learn ͷؔ਺΍ΫϥεΛ import
     let load_digits = Python.import("sklearn.datasets").load_digits
     let LinearSVC = Python.import("sklearn.svm").LinearSVC
     let train_test_split = Python.import("sklearn.model_selection").train_test_split
     // ֶश
     let classifier = LinearSVC()
     let dataset = load_digits()
     let (X_train, X_test, y_train, y_test)
     = train_test_split(dataset["data"], dataset["target"]).tuple4
     classifier.fit(X_train, y_train)
     // ධՁ
     print("train: \(classifier.score(X_train, y_train))")
     print("test: \(classifier.score(X_test, y_test))")
     

     View Slide

131. ࣮ߦͷ༷ࢠ
     

     View Slide

132. !
     ϞσϧΛ Core ML ༻ʹ
     ม׵ͯ͠ iOS ΞϓϦͰ࢖͏
     

     View Slide

133. coremltools 6
     6 h$ps:/
     /github.com/apple/coremltools
     

     View Slide

134. Core ML ༻ͷϞσϧΛ࡞੒
     // coremltools ͷ import
     let coremltools = Python.import("coremltools")
     // Ϟσϧͷม׵ͱอଘ
     let coreml_model = coremltools.converters.sklearn
     .convert(classifier)
     coreml_model.save("Digits.mlmodel")
     

     View Slide

135. coremltools Λ import
     // coremltools ͷ import
     let coremltools = Python.import("coremltools")
     // Ϟσϧͷม׵ͱอଘ
     let coreml_model = coremltools.converters.sklearn
     .convert(classifier)
     coreml_model.save("Digits.mlmodel")
     

     View Slide

136. ϞσϧΛม׵
     // coremltools ͷ import
     let coremltools = Python.import("coremltools")
     // Ϟσϧͷม׵ͱอଘ
     let coreml_model = coremltools.converters.sklearn
     .convert(classifier)
     coreml_model.save("Digits.mlmodel")
     

     View Slide

137. ϞσϧΛอଘ
     // coremltools ͷ import
     let coremltools = Python.import("coremltools")
     // Ϟσϧͷม׵ͱอଘ
     let coreml_model = coremltools.converters.sklearn
     .convert(classifier)
     coreml_model.save("Digits.mlmodel")
     

     View Slide

138. View Slide

139. View Slide

140. iOS ΞϓϦ͔Βม׵ͨ͠ϞσϧΛར༻
     class ViewController: UIViewController {
     @IBOutlet var canvasView: CanvasView!
     private let classifier = Digits()
     @IBAction func pressClassifyButton(_ button: UIButton) {
     let image = Image(uiImage: canvasView.image).resizedTo(width: 8, height: 8)
     let input = try! MLMultiArray(shape: [8, 8], dataType: .double)
     var pointer = input.dataPointer.bindMemory(to: Double.self, capacity: 8 * 8)
     for pixel in image {
     pointer.pointee = Double(255 - pixel) / 16.0
     pointer += 1
     }
     let result = try! classifier.prediction(input: DigitsInput(input: input))
     let alertController = UIAlertController(title: "\(result.classLabel)", message: nil, preferredStyle: .alert)
     alertController.addAction(UIAlertAction(title: "Dismiss", style: .default) { [weak self] _ in
     self?.canvasView.clear()
     })
     present(alertController, animated: true, completion: nil)
     }
     }
     

     View Slide

141. ෼ྨثͷੜ੒
     class ViewController: UIViewController {
     @IBOutlet var canvasView: CanvasView!
     private let classifier = Digits()
     @IBAction func pressClassifyButton(_ button: UIButton) {
     let image = Image(uiImage: canvasView.image).resizedTo(width: 8, height: 8)
     let input = try! MLMultiArray(shape: [8, 8], dataType: .double)
     var pointer = input.dataPointer.bindMemory(to: Double.self, capacity: 8 * 8)
     for pixel in image {
     pointer.pointee = Double(255 - pixel) / 16.0
     pointer += 1
     }
     let result = try! classifier.prediction(input: DigitsInput(input: input))
     let alertController = UIAlertController(title: "\(result.classLabel)", message: nil, preferredStyle: .alert)
     alertController.addAction(UIAlertAction(title: "Dismiss", style: .default) { [weak self] _ in
     self?.canvasView.clear()
     })
     present(alertController, animated: true, completion: nil)
     }
     }
     

     View Slide

142. View Slide

143. Digits.mlmodel ͔Β෼ྨثͷΫϥε͕ࣗಈੜ੒͞ΕΔ
     class Digits {
     var model: MLModel
     convenience init() { ... }
     func prediction(input: DigitsInput) throws -> DigitsOutput {
     let outFeatures = try model.prediction(from: input)
     let result = DigitsOutput(
     classLabel: outFeatures.featureValue(for: "classLabel")!.int64Value,
     classProbability: outFeatures.featureValue(
     for: "classProbability"
     )!.dictionaryValue as! [Int64 : Double]
     )
     return result
     }
     ...
     }
     

     View Slide

144. Ϙλϯ͕ԡ͞Εͨͱ͖
     class ViewController: UIViewController {
     @IBOutlet var canvasView: CanvasView!
     private let classifier = Digits()
     @IBAction func pressClassifyButton(_ button: UIButton) {
     let image = Image(uiImage: canvasView.image).resizedTo(width: 8, height: 8)
     let input = try! MLMultiArray(shape: [8, 8], dataType: .double)
     var pointer = input.dataPointer.bindMemory(to: Double.self, capacity: 8 * 8)
     for pixel in image {
     pointer.pointee = Double(255 - pixel) / 16.0
     pointer += 1
     }
     let result = try! classifier.prediction(input: DigitsInput(input: input))
     let alertController = UIAlertController(title: "\(result.classLabel)", message: nil, preferredStyle: .alert)
     alertController.addAction(UIAlertAction(title: "Dismiss", style: .default) { [weak self] _ in
     self?.canvasView.clear()
     })
     present(alertController, animated: true, completion: nil)
     }
     }
     

     View Slide

145. ෼ྨثʹ౉͢ೖྗ஋Λੜ੒
     class ViewController: UIViewController {
     @IBOutlet var canvasView: CanvasView!
     private let classifier = Digits()
     @IBAction func pressClassifyButton(_ button: UIButton) {
     let image = Image(uiImage: canvasView.image).resizedTo(width: 8, height: 8)
     let input = try! MLMultiArray(shape: [8, 8], dataType: .double)
     var pointer = input.dataPointer.bindMemory(to: Double.self, capacity: 8 * 8)
     for pixel in image {
     pointer.pointee = Double(255 - pixel) / 16.0
     pointer += 1
     }
     let result = try! classifier.prediction(input: DigitsInput(input: input))
     let alertController = UIAlertController(title: "\(result.classLabel)", message: nil, preferredStyle: .alert)
     alertController.addAction(UIAlertAction(title: "Dismiss", style: .default) { [weak self] _ in
     self?.canvasView.clear()
     })
     present(alertController, animated: true, completion: nil)
     }
     }
     

     View Slide

146. ෼ྨ
     class ViewController: UIViewController {
     @IBOutlet var canvasView: CanvasView!
     private let classifier = Digits()
     @IBAction func pressClassifyButton(_ button: UIButton) {
     let image = Image(uiImage: canvasView.image).resizedTo(width: 8, height: 8)
     let input = try! MLMultiArray(shape: [8, 8], dataType: .double)
     var pointer = input.dataPointer.bindMemory(to: Double.self, capacity: 8 * 8)
     for pixel in image {
     pointer.pointee = Double(255 - pixel) / 16.0
     pointer += 1
     }
     let result = try! classifier.prediction(input: DigitsInput(input: input))
     let alertController = UIAlertController(title: "\(result.classLabel)", message: nil, preferredStyle: .alert)
     alertController.addAction(UIAlertAction(title: "Dismiss", style: .default) { [weak self] _ in
     self?.canvasView.clear()
     })
     present(alertController, animated: true, completion: nil)
     }
     }
     

     View Slide

147. ݁ՌΛ UIAlertController Ͱදࣔ
     class ViewController: UIViewController {
     @IBOutlet var canvasView: CanvasView!
     private let classifier = Digits()
     @IBAction func pressClassifyButton(_ button: UIButton) {
     let image = Image(uiImage: canvasView.image).resizedTo(width: 8, height: 8)
     let input = try! MLMultiArray(shape: [8, 8], dataType: .double)
     var pointer = input.dataPointer.bindMemory(to: Double.self, capacity: 8 * 8)
     for pixel in image {
     pointer.pointee = Double(255 - pixel) / 16.0
     pointer += 1
     }
     let result = try! classifier.prediction(input: DigitsInput(input: input))
     let alertController = UIAlertController(title: "\(result.classLabel)", message: nil, preferredStyle: .alert)
     alertController.addAction(UIAlertAction(title: "Dismiss", style: .default) { [weak self] _ in
     self?.canvasView.clear()
     })
     present(alertController, animated: true, completion: nil)
     }
     }
     

     View Slide

148. Swi$Digits 7
     7 h$ps:/
     /github.com/koher/swi6-digits
     

     View Slide

149. ࣮ࡍͷͱ͜Ζ
     - ϓϩμΫγϣϯ؀ڥͰ࢖͏ͷ͸·ͩා͍
     - ֶश͸ਪ࿦ΑΓ΋ະख़ͳٕज़Λ࢖͍΍͍͢
     - Python Λ஌Βͳͯ͘ྑ͍Θ͚Ͱ͸ͳ͍
     - Swift Ͱ΋ॻ͚ΔॲཧΛ Swift Ͱॻ͚Δͷ͸͏Ε͍͠
     - Swift ͷະདྷΛࠓ͔ΒମݧͰ͖Δ
     

     View Slide

150. !
     ࣗ෼ͰֶͿͱ͖ʹ໾ཱͭࢿྉ
     

     View Slide

151. Machine Learning | Coursera 8
     8 h$ps:/
     /www.coursera.org/learn/machine-learning
     

     View Slide

152. PythonͰ͸͡ΊΔػցֶश 9
     9 h$ps:/
     /www.oreilly.co.jp/books/9784873117980/
     

     View Slide

153. ѻΘͳ͔ͬͨτϐοΫ
     -
     !
     ݕূσʔλʢ Validation data ʣ
     -
     !
     લॲཧʢ Preprocessing ʣ
     -
     !
     ਖ਼ଇԽʢ Regularization ʣ
     -
     !
     ަࠩݕূʢ Cross-validation ʣ
     -
     !
     ِཅੑʢ False Positive ʣɺِӄੑʢ False Negative ʣ
     -
     !
     ద߹཰ʢ Precision ʣɺ࠶ݱ཰ʢ Recall ʣɺ
     ɹ F ஋ʢ F-measure ʣ
     - ...
     

     View Slide

154. View Slide

155. Ϟσϧͷબ୒ʹςετσʔλΛར༻ͯ͠
     ਖ਼͘͠ධՁͰ͖͍ͯΔͷ͔ʁ
     

     View Slide

156. !
     ݕূσʔλʢ Valida'on data ʣ
     ͰϞσϧΛධՁͯ͠બ୒͠ɺ
     ςετσʔλͰ൚ԽೳྗΛධՁ͢Δ
     

     View Slide

157. SE-0195: @dynamicMemberLookup 2
     2 h$ps:/
     /github.com/apple/swi6-evolu9on/blob/master/proposals/0195-dynamic-member-lookup.md
     

     View Slide

158. SE-0216: @dynamicCallable 3
     3 h$ps:/
     /github.com/apple/swi6-evolu9on/blob/master/proposals/0216-dynamic-callable.md
     

     View Slide

159. Swi$ for TensorFlow 4
     4 h$ps:/
     /medium.com/tensorflow/introducing-swi8-for-tensorflow-b75722c58df0
     

     View Slide

160. Swi$ for TensorFlow ͷΠϯετʔϧ 10
     10 h%ps:/
     /github.com/tensorflow/swi8/blob/master/Installa

     View Slide

161. Xcode 10 beta Λ࢖͏৔߹
     sudo xcode-select -s /Applications/Xcode-beta.app/
     

     View Slide

162. Core ML ༻ʹϞσϧΛม׵ 11
     11 h$ps:/
     /developer.apple.com/documenta6on/coreml/conver6ngtrainedmodelstocore_ml
     

     View Slide

163. coremltools 6
     6 h$ps:/
     /github.com/apple/coremltools
     

     View Slide

164. ͓͠·͍
     

     View Slide