Big Data, Predictive Modeling and tools

CCIP 2015

Transcript

1. Big Data, Predictive Modeling & Tools Olivier Grisel — @ogrisel CCIP October 2015
2. About me
3. Big Data as a buzzword
4. Regionator 3000 http://labs.data-publica.com/regionator3000/
5. http://transports.blog.lemonde.fr/2014/06/05/regionator-la- carte-de-france-dessinee-par-les-trajets-quotidiens/
6. http://transports.blog.lemonde.fr/2014/06/05/regionator-la- carte-de-france-dessinee-par-les-trajets-quotidiens/
7. http://www.insee.fr/fr/themes/detail.asp? reg_id=99&ref_id=mobilite-professionnelle-10
8. http://www.insee.fr/fr/themes/detail.asp? reg_id=99&ref_id=mobilite-professionnelle-10
9. 120% of the capacity of a 3’5 floppy disk
10. Big Data ≠ Predictive Analytics
11. Outline • What Big Data actually is or isn’t • Predictive modeling concepts & tools • Big Data architecture & tools
12. How big is Big Data?
13. 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
14. Reading the full content of a 1TB HDD at 100MB/s: 2 hours 45 minutes
15. Canonical Big Data problem: indexing the Web • For each word, aggregate the list of page URLs • 2 billion HTML pages: 100TB >10 days just to read sequentially
18. Other Big Data examples • GSM location event log from telco • Transaction log of a big retail network • Activity records for a service with 10s of millions of users
19. What is Predictive Analytics? and Machine Learning
20. 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
21. 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 ?
22. 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
23. Training text docs images sounds transactions Labels Machine Learning Algorithm Model Predictive Modeling Data Flow Feature vectors
24. 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
25. Predictive Models • 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
26. Tools for predictive analytics
27. SPSS MATLAB
28. SPSS MATLAB
29. 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
30. 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
31. 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
32. 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
35. 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
36. 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
37. 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
38. 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
39. Big Data architecture(s) & tools
40. Distributed Event Log & Queues Distributed Event Stream Processing Distributed Storage Distributed Batch Processing On-line Transaction Processing & App Views Analytical Database Predictive Models
43. Thank you! Questions? @Inria @ogrisel http://scikit-learn.org
44. Bonus tracks
45. Back to the Regionator What if we did not have census data on daily mobility?
46. 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
47. Predictive Analytics on Big Data
49. BIG DATA
50. BIG DATA small(er) data
51. BIG DATA small(er) data
52. 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
53. Data size and modeling quality
54. – Peter Norvig, Research Director, Google “We don’t have better algorithms. We just have more data.”
58. More data beats better models?
59. http://technocalifornia.blogspot.fr/2012/07/more-data-or-better- models.html
60. Let’s train a parametric model to read handwritten digits from gray level pixels.
62. model stops improving
64. Bias vs Variance
65. high bias
66. high bias high variance
67. high bias high variance low variance
68. Variance solution #1: collect more samples
69. Let’s train a non-parametric model to read handwritten digits from gray level pixels.
71. high variance almost no bias variance decreasing with #samples
72. Bias solution #1: non-parametric models
73. Type # rooms Surface (m2) Floor Public Transp. Apart. 3 65 2 Yes House 5 110 NA No Duplex 4 95 4 Yes features samples
74. 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
75. Data has 2 dimensions: # samples and # features
76. Bias solution #2: richer features
78. • 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
79. • 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