Big Data and Predictive Modeling

Web We Can 2015

Transcript

1. Big Data and Predictive Modeling Olivier Grisel — @ogrisel Web We Can March 21, 2015
2. About me
3. Big Data as a buzzword
4. Triumph of the Nerds: Nate Silver Wins in 50 States http://mashable.com/2012/11/07/nate-silver-wins/
5. Triumph of the Nerds: Nate Silver Wins in 50 States http://mashable.com/2012/11/07/nate-silver-wins/
6. Nate Silver’s election model, Big Data? $ git clone gh:jseabold/538model $ du -h 538model/data 188K 538model/data
7. 15% of the capacity of a 3’5 floppy disk
8. Regionator 3000 http://labs.data-publica.com/regionator3000/
9. http://transports.blog.lemonde.fr/2014/06/05/regionator-la- carte-de-france-dessinee-par-les-trajets-quotidiens/
10. http://transports.blog.lemonde.fr/2014/06/05/regionator-la- carte-de-france-dessinee-par-les-trajets-quotidiens/
11. http://www.insee.fr/fr/themes/detail.asp? reg_id=99&ref_id=mobilite-professionnelle-10
12. http://www.insee.fr/fr/themes/detail.asp? reg_id=99&ref_id=mobilite-professionnelle-10
13. 120% of the capacity of a 3’5 floppy disk
14. Big Data ≠ Predictive Analytics
15. Predictive Analytics ≠ Descriptive Analytics
16. Goals of this talk • What Big Data actually is or isn’t • Introduce predictive modeling concepts • Contrast predictive analytics vs descriptive analytics
17. How big is Big Data?
18. – Wikipedia “Big data is a blanket term for any collection of data sets so large and complex that it becomes difficult to process using on-hand database management tools or traditional data processing applications.”
19. Not Big Data • Data that fits on a spreadsheet • Data that can be analyzed in RAM (< 10 GB) • Data operations that can be performed quickly by a traditional database, e.g. single node PostgreSQL server
20. Reading the full content of a 1TB HDD at 100MB/s: 2 hours 45 minutes
21. Canonical Big Data problem: indexing the Web • Inverted index on tera bytes of text data • Process each HTML page as a URL + bag of words • For each word, aggregate the list of page URLs • 2 billion HTML pages: 100TB >10 days just to read sequentially
25. Other Big Data examples • GSM location event log from telco • Transaction log of a big retail network • Raw traffic data on a large website • Activity records for a service with 10s of millions of users
26. Not Big Data • Polls data (~10K data points) • Census data (~10M data points) • Real estate transactions data (~10M data points) • Any dataset publicly available for download
27. What is Predictive Analytics? and Machine Learning
28. Type # rooms Surface m2 Floor Public Transports Apartment 3 65 2 Yes House 5 110 NA No Duplex 4 95 4 Yes Sold 300k 1.5M 2.2M
29. Type # rooms Surface m2 Floor Public Transports Apartment 3 65 2 Yes House 5 110 NA No Duplex 4 95 4 Yes features samples Sold 300k 1.5M 2.2M target
30. Type # rooms Surface m2 Floor Public Transports Apartment 3 65 2 Yes House 5 110 NA No Duplex 4 95 4 Yes features samples Sold 300k 1.5M 2.2M target Apartment 2 35 3 Yes ?
31. Predictive Modeling • Automated predictions of outcome on new data • Alternative to hard-coded rules written by experts • Extract the structure of historical data • Statistical tools to summarize the training data into an executable predictive model
32. Training text docs images sounds transactions Labels Machine Learning Algorithm Model Predictive Modeling Data Flow Feature vectors
33. New text doc image sound transaction Model Expected Label Predictive Modeling Data Flow Feature vector Training text docs images sounds transactions Labels Machine Learning Algorithm Feature vectors
34. Descriptive vs Predictive
35. Descriptive Statistics • Ex: Sales by (day, months, year) x region • Graphical visualization: get insights, tell a story to explain what’s happening in the data • Realm of Business Intelligence: reports & dashboard for managers • A wrong decision can be very costly • Small number of important decisions made by a human
36. Predictive Statistics • Ex: Movie recommendations or targeted ads • Embedded in a user service to make it more useful / attractive / profitable • A wrong individual decision is not costly • Large number of small automated decisions • Humans would not be fast enough to make the predictions
37. Mixed models • Predictive modeling to identify interesting subsets of the data • Ex: Fraud detection, churn forecasting • Help human decision makers focus on important cases • Human expert feedback to improve predictive models
38. Key takeaway points:
39. • Big Data ≠ Predictive Analytics • Predictive Analytics • Automated decision making embedded in products (e.g. recommenders) • Individual bad decisions are typically not costly • Descriptive Analytics • Business Intelligence: human decision making • Individual bad decisions can be very costly
40. Thank you! Questions? @Inria @ogrisel http://scikit-learn.org
41. Bonus tracks
42. Back to the Regionator What if we did not have census data on daily mobility?
43. Back to the Regionator • Use raw daily telco logs • Group By (phone, day) to extract daily trips • Join By GPS coordinates to “departement” names • Filter out small trips • Group By (home, work) “departements” • Count
44. Tools for predictive analytics
45. SPSS MATLAB
46. SPSS MATLAB
47. New text doc image sound transaction Model Expected Label Small data Training text docs images sounds transactions Labels Machine Learning Algorithm Feature vectors Feature vector
48. New text doc image sound transaction Model Expected Label Small / Medium data Training text docs images sounds transactions Labels Machine Learning Algorithm Feature vectors Feature vector
49. New text doc image sound transaction Model Expected Label Small / Medium data with Training text docs images sounds transactions Labels Machine Learning Algorithm Feature vectors Feature vector
50. New text doc image sound transaction Model Expected Label Small / Medium data with Training text docs images sounds transactions Labels Machine Learning Algorithm Feature vectors Feature vector
53. Predictive Analytics on Big Data
54. Model Expected Label Big data with Machine Learning Algorithm New text doc image sound transaction Training text docs images sounds transactions Labels Feature vectors Feature vector
55. Model Expected Label Big data with Machine Learning Algorithm New text doc image sound transaction Training text docs images sounds transactions Labels Feature vectors Feature vector
56. Model Expected Label Big data with Machine Learning Algorithm New text doc image sound transaction Training text docs images sounds transactions Labels Feature vectors Feature vector
58. BIG DATA
59. BIG DATA small(er) data
60. BIG DATA small(er) data
61. From Big to Small • Feature extraction often shrinks data • Filter / Join / Group By / Count • Machine Learning performed on aggregates • Sampling for fast in-memory iterative modeling
62. Data size and modeling quality
63. – Peter Norvig, Research Director, Google “We don’t have better algorithms. We just have more data.”
67. More data beats better models?
68. http://technocalifornia.blogspot.fr/2012/07/more-data-or-better- models.html
69. Let’s train a parametric model to read handwritten digits from gray level pixels.
71. model stops improving
73. Bias vs Variance
74. high bias
75. high bias high variance
76. high bias high variance low variance
77. Variance solution #1: collect more samples
78. Let’s train a non-parametric model to read handwritten digits from gray level pixels.
80. high variance almost no bias variance decreasing with #samples
81. Bias solution #1: non-parametric models
82. Type # rooms Surface (m2) Floor Public Transp. Apart. 3 65 2 Yes House 5 110 NA No Duplex 4 95 4 Yes features samples
83. Type # rooms Surface (m2) Floor Public Transp. School (km) Flood plain Apart. 3 65 2 Yes 1.0 No House 5 110 NA No 25.0 Yes Duplex 4 95 4 Yes 0.5 No features samples
84. Bias solution #2: richer features
85. Data has 2 dimensions: # samples and # features
87. • Parametric e.g. linear model (traditional stats) vs Non-parametric e.g. Random Forests, Neural Networks (Machine Learning) • Understand a model with 10% accuracy vs blindly trust a model with 90% accuracy • Simple models e.g. F = m a, F = - G (m1 + m2) / r^2 will not become false(r) because of big data • New problems can be tackled: computer vision, speech recognition, natural language understanding
88. • the (experimental) scientific method introduced by Karl Popper is based on the falsifiability of formulated hypotheses • theory is correct as long as past predictions hold in new experiments • machine learning train-validation-test splits and cross-validation is similar in spirit • ml model is just a complex theory: correct as long as its predictions still hold