Exploratory Seminar: An Introduction to K-Means Clustering

This is to introduce K-Means Clustering algorithm, which segments the data based on a given set of variables, by demonstrating it with Exploratory’s Analytics view.

Transcript

1. EXPLORATORY
2. Kan Nishida co-founder/CEO Exploratory Summary Beginning of 2016, launched Exploratory, Inc. to make Data Science available for everyone. Prior to Exploratory, Kan was a director of development at Oracle leading development teams for building various Data Science products in areas including Machine Learning, BI, Data Visualization, Mobile Analytics, Big Data, etc. While at Oracle, Kan also provided training and consulting services to help organizations transform with data. @KanAugust Instructor
3. Mission Make Data Science available for everyone
4. Data Science is not just for Engineers and Statisticians. Exploratory makes it possible for Everyone to do Data Science. The Third Wave
5. First Wave Second Wave Third Wave Proprietary Open Source UI & Programming Programming 2016 2000 1976 Monetization Commoditization Democratization Statisticians Data Scientists Smart Waves - Machine Learning / AI Algorithms Experience Tools Open Source UI & Automation Business Users Theme Users Exploratory
6. Questions Data Science Workflow Communication Data Access Data Wrangling Data Visualization Machine Learning / Statistics Exploration
7. Questions What you can do with Exploratory Communication Data Access Data Wrangling Visualization Machine Learning / Statistics Exploratory Data Analysis
8. Questions Communication Data Access Data Wrangling Visualization Machine Learning / Statistics Exploratory Data Analysis
9. K-Means Clustering
10. • Supervised Learning • Unsupervised Learning Machine Learning
11. • Supervised Learning • Unsupervised Learning Machine Learning
12. • Detect patterns / trends in Data and produce insights. • You don’t have answers in your Data. Unsupervised Learning
13. Unsupervised Learning • Clustering • Anomaly Detection • MDS (Multi-Dimensional Scaling) • PCA / SVD (Dimensionality Reduction)
14. Unsupervised Learning • Clustering • Anomaly Detection • MDS (Multi-Dimensional Scaling) • PCA / SVD (Dimensionality Reduction)
15. Unsupervised Learning • Clustering • Anomaly Detection • MDS (Multi-Dimensional Scaling) • PCA / SVD (Dimensionality Reduction)
16. Clustering
17. Clustering
18. Attributes Woman Generation X Actor High Income American Southern Long Hair
19. Also known as Variables, Features, Predictors, etc. Woman Generation X Actor High Income American Southern Long Hair
21. Politicians Actors
23. Not yet… Getting accused of…
24. • K-Means • Hierarchical Clustering • GMM (Gaussian mixture model) • LDA Algorithms
25. • K-Means • Hierarchical Clustering • GMM (Gaussian mixture model) • LDA Algorithms
26. K-Means • Specify the number of clusters • Only based on numeric variables • Can be affected by noise and outliers
27. Example
28. US Baby Data
31. Cluster US States into a few groups based on their similarity using Father Age and Mother Age variables.
32. Data Preparation Each subjects (US State) needs to be presented as its own unique row. One row is one state.
35. 1. Group Data by US State 2. Summarize - Mean of Mother Age and Mean of Father Age Data Preparation
36. Group the data by US State. From the column menu at the “state” column (either on Table view or Summary view), select “Group By” 1. Group By
37. Select “Summarize (Aggregate)”, then select “mean (Average)” from the column header menu for Mother Age. 2.1. Summarize for Mean of Mother Age
38. 2.2. Summarize for Mean of Father Age
39. Summarized data at US States level.
40. Run K-Means Clustering!
41. How K-Means Algorithm Works?
44. 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
45. 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
46. 46 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
47. 47 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
48. 48 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
49. 49 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
50. 50 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
51. 51 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
52. 52 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
53. 53 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
54. 54 1. Set the center of clusters randomly. 2. Each dot should belong to the nearest center of the cluster. 3. Move each center of clusters to the average of its members. 4. Repeat the step2 and step3 until all centers don’t move anymore.
55. • Command Step • Analytics View K-Means Clustering in Exploratory
56. • Command Step • Analytics View K-Means Clustering in Exploratory
57. • Select mother_age_mean, father_age_mean with Control (Command) + Click. • Click the column menu, select “Analytics”, select“K-Means” and select “Selected Columns”. Run K-Means Clustering
58. Now the US States are clustered (grouped) based on the mother and father ages. Each US State now has the cluster ID.
59. Visualize it!
60. Scatter
63. Map
65. It doesn’t have to be just two variables. You can use many variables.
66. Let’s add another variable “Cigarette Use Ratio” as one of the clustering variables. Add Cigarette Use Ratio
67. 1. Create Cigarette Use Ratio column 2. Add Cigarette Use Ratio column to K- Means and re-run. Create Cigarette Use Ratio Column
68. Summarize with ‘Ratio of TRUE’ 1. Create Cigarette Use Ratio Column
69. Summarize with ‘Ratio of TRUE’
70. Summarize with ‘Ratio of TRUE’
71. Select ‘K-Means’ step and Click the token to add cigarette_use_ratio column. Add cigarette_use_ratio column and re-run K-Means
72. Add ‘cigarette_use_ratio’ column. Add cigarette_use_ratio column and re-run K-Means
73. Clusters are recalculated. Add cigarette_use_ratio column and re-run K-Means
74. Opening the chart again will automatically refresh and use the new clusters. But, it looks the same as before… why?
75. The clusters are separated by mother_age_mean variable, but not by cigarette_use_ratio variable.
76. Compared to mother_age and father_age, cigarette_use_ratio values are very small.
77. K-Means Algorithm • It calculates the distances between the centers of the clusters and the members. So the variables with larger values tend to have bigger influence on how the data get clustered. • Mother Age and Father Age values are in a similar range while cigarette_use_ratio values are much smaller than the ages.
78. Normalize!
79. Select all three columns, and select ’normalize’ function from the column header menu.
80. Values are normalized in each variable.
81. We can compare the variables with a same scale.
82. Cluster 2 and 3 can not be separated only by Mother and Father ages. This means Cigarette Use Ratio makes the difference.
83. Cluster 2 and 3 are differentiated by Cigarette Use Ratio values.
84. Using Analytics View
85. K-Means with Analytics View • It builds a K-Means Clustering Model and generate a set of pre- defined charts to help you understand the characters of the clusters. • It normalizes the data before building the model.
86. Select ‘K-Means Clustering’
87. Select all 3 numerical columns.
89. • Each dot represents each US State. • X and Y Axis are artificially created dimensions to express as much of the variance of the original measures (Mother Age, Father Age, Cigarette Use Ratio)
90. • The original measures (Mother Age, Father Age, Cigarette Use Ratio) are shown as gray lines. • The dots closer to the gray lines have higher values on those lines. • Example: Blue cluster tend to have large numbers of Father Age and Mother Age.
92. • Cluster 1: Father Age and Mother Age are higher. Cigarette Use Ratio is lower. • Cluster 2: Father Age, Mother Age, and Cigarette Use Ratio are lower. • Cluster 3: Father Age and Mother Age are lower. Cigarette Use Ratio is higher.
93. Visualizing K-Means Clustering Results to Understand the Characteristics of Clusters Better Take a look at the following blog post for more details. (https://exploratory.io/note/kanaugust/Introduction-to-K-Means-Clustering-under-Analytics-View-bjW2EZc3Ge)
94. How many clusters should we build?
95. Elbow Curve Method
96. Select TRUE for Elbow Method option.
97. The distances between the center and the members are shown at Y-Axis. The point where the value drops significantly means that increasing the number of the clusters is contributing to capture the differences among the data in a meaningful way.
98. Future Seminars
99. January 8th (Tuesday), 2019 • Data Wrangling: Working with Date / Time Data and Visualize It Planned • Analytics 101 - When to use which algorithms? • Data Wrangling: Introduction to Regular Expression https://exploratory.io/online-seminar
100. Contact Email kan@exploratory.io Data Science Training https://exploratory.io/training Twitter @KanAugust Online Seminar https://exploratory.io/online-seminar