Machine Learning

Transcription

1 Machine Learning

2

3

4

5 John Jane Miko Lee

6 Job Postings for Machine Learning Source: Indeed.com

7 Average Salary by Job Type (USA) Source: Stack Overflow 2017

8

9 Overview 1. Introduction to ML 2. Introduction to R 3. Classification 4. Regression 5. Beyond the Basics

10 About Me Data Science Consultant Education B.S. in Computer Science B.A. in Philosophy Community Public Speaker Pluralsight Author Microsoft MVP ASPInsider Open-source Software

11 How Does This Apply to Me? Make decisions using data Make predictions using data Make recommendations using data Find patterns of interest in data Find anomalies in data Write code that does all these things

12

13

14 Introduction to Machine Learning

15 What is Machine Learning?

16 Artificial Intelligence Machine Learning Statistics

17 f x

18 f x Data Function Prediction

19 f x Data Function Prediction

20 f x Data Function Prediction Cat Dog

21 f x Data Function Prediction Cat Dog

22 f x Data Function Prediction Cat Dog Is cat?

23 f x Data Function Prediction Cat Dog Is cat?

24 f x Data Function Prediction Cat Dog Is cat? Yes

25 Find a question

26 Find a question Prepare the data

27 Find a question Prepare the data Train the model

28 Find a question Prepare the data Train the model Evaluate the model

29 Find a question Prepare the data Deploy the model Train the model Evaluate the model

30 Find a question Monitor the model Prepare the data Deploy the model Train the model Evaluate the model

31 Find a question Monitor the model Prepare the data Deploy the model Train the model Evaluate the model

32 Data

33 Training Data Test

34 ML Algorithm Training Data Test

35 ML Algorithm ML Model Training Data Test

36 ML Algorithm ML Model Training Data Test

37 ML Algorithm ML Model Training Data Test New Data

38 ML Algorithm ML Model Training Prediction Data Test New Data

39 What Can Machine Learning Do?

40 f x 1.23

41 Source: Futurama

42 Introduction to R

43 What is R? Open source Language and environment Numerical and graphical analysis Cross platform

44 What is R? Active development Large user community Modular and extensible extensions

45 FREE

46 FREE

47

48

49

50 Code Demo

51 Classification

52 f x

53 Count of Spam Words Correct Spelling Ratio

54 Count of Spam Words Correct Spelling Ratio

55 Count of Spam Words Correct Spelling Ratio

56 Count of Spam Words Correct Spelling Ratio

57 Count of Spam Words Correct Spelling Ratio

58 Count of Spam Words Correct Spelling Ratio

59 Classification Algorithms k-nearest Neighbor Classifier Decision Tree Classifier Naïve Bayes Classifier Support Vector Machine Neural Network Classifier x2 x1

60 Decision Tree Classifier Supervised learning is age > 9.5? is sex male? Survived Died is family > 2.5? Died Survived

61 Decision Tree Classifier is sex male? Supervised learning Tree of decisions is age > 9.5? Survived Died is family > 2.5? Died Survived

62 Decision Tree Classifier is sex male? Supervised learning Tree of decisions Information gain Died is age > 9.5? is family > 2.5? Survived Died Survived

63 Decision Tree Classifier is sex male? Supervised learning Tree of decisions Information gain Simple and easy Died is age > 9.5? is family > 2.5? Survived Died Survived

64 Titanic Passenger Manifest Name Gender Age Family Survived Elizabeth Allen Female 29 0 Yes Hudson Allison Jr. Male 1 3 Yes Helen Allison Female 2 3 No Hudson Allison Sr. Male 30 3 No Bessie Allison Female 25 3 No

65 is sex male? is age > 9.5? Survived Died is family > 2.5? Died Survived

66 Neural Network Classifier Supervised learning Source: Wikipedia

67 Neural Network Classifier Supervised learning Neurons in a brain Source: Wikipedia

68 Neural Network Classifier Supervised learning Neurons in a brain Complex Source: Wikipedia

69 Neural Network Classifier Supervised learning Neurons in a brain Complex Not transparent Source: Wikipedia

70 Real-World Examples Should we approve this loan? Will this customer buy from us? Should we replace this part? Does this person have cancer? x2 x1

71 Iris Data Set Iris Setosa Iris Versicolor Iris Virginica Photos by Radomił Binek, Danielle Langlois, and Frank Mayfield

72 Iris Data Set Fisher s Iris Data Species Petal Length Petal Width Sepal Length Sepal Width setosa setosa setosa setosa setosa

73 Classification Demo Goal: Predict species based on petal and sepal measurements

74 Regression

75 f x 1.23

76 Sale Price Area

77 Sale Price Area

78 Sale Price Area

79 Regression Algorithms Linear Regression Polynomial Regression Lasso Regression ElasticNet Regression Neural Network Regression x2 x1

80 Simple Linear Regression Relationship

81 Simple Linear Regression Relationship Linear model

82 Simple Linear Regression Relationship Linear model Explanatory variable

83 Simple Linear Regression Relationship Linear model Explanatory variable Outcome variable

84 Simple Linear Regression Linear predictor function

85 Simple Linear Regression Linear predictor function y = m x + b

86 Simple Linear Regression Linear predictor function y = m x + b Parameters estimated

87 Simple Linear Regression Linear predictor function y = m x + b Parameters estimated Relies on assumptions

88 Neural Network Regression Same as before Numeric vs. Categorical Source: Wikipedia

89 Real-World Examples How much profit will we make? What will the price be tomorrow? How many will this person buy? How long until this part fails? x2 x1

90 Regression Demo Goal: Predict petal width of Iris flowers

91 Beyond the Basics

92 This is just the tip of the iceberg! This is just the tip of the iceberg!

93 Find a question Monitor the model Prepare the data Deploy the model Train the model Evaluate the model

94 Creating accurate and robust models is not easy

95 Find a question Monitor the model Prepare the data Deploy the model Train the model Evaluate the model

96 Data are messy Cleaning and Transforming Data

97 Cleaning and Transforming Data Data are messy 80% of work

98 Cleaning and Transforming Data Data are messy 80% of work R helps a lot

99 Cleaning and Transforming Data Data are messy 80% of work R helps a lot Record all steps

100 Goodness of Fit

101 Underfit Goodness of Fit

102 Goodness of Fit Underfit Overfit

103 Goodness of Fit Underfit Good fit Overfit

104 Deep Learning

105 John Jane Miko Lee

106

107 f x 1.23

108 f x 1.23

109 f x 1.23

110 Source: YOLO: Real-Time Object Detection

111 Source: Source: Nvidia

112

113 Source:

114 Source: Pouff Google - Grocery Deep Mind Trip

115 Source: Boston Dynamics

116 Practical Demo Goal: Predict who will survive the Titanic

117 Conclusion

118 Where to Go Next Pluralsight: Data Camp: Coursera: Tensorflow:

119

120

121 Feedback Very important to me! What did you like? What could I improve?

122 Conclusion 1. Introduction to ML 2. Introduction to R 3. Classification 4. Regression 5. Beyond the Basics

123

124 Are you prepared? Is your organization? Is our world prepared?

125

126 Contact Info Matthew Renze Data Science Consultant Renze Consulting Website: Thank You! : )