Machine Learning: A practical approach

These are the slides for the presentation of my B.Sc. CS thesis: https://github.com/josemazo/bsc-cs-thesis

All the images I use in the slides are owned by their creators. Here it's a list their sources:

IBM Watson
https://www.flickr.com/photos/ascentstage/5471363235
Books
https://www.flickr.com/photos/meeli/2854849909
WhatsApp
http://www.techaudible.net/enable-disable-last-seen-whatsapp-plus-whatsapp/
Travel salesman problem
http://eff-stoplocal.blogspot.com.es/2010/11/sepia-saturday-finding-lost-treasure-pt_12.html
Engineering students
https://www.flickr.com/photos/pellethepoet/12097798504/
Bridges Konigsberg
https://people.kth.se/~carlofi/teaching/FEL3250-2013/courseinfo.html
Bruce Lee
http://en.wikipedia.org/wiki/Bruce_Lee
BMO
http://lightningjinx.deviantart.com/art/Adventure-Time-BMO-350560854
http://es.horadeaventura.wikia.com/wiki/BMO
Obama
https://www.targetcomponents.co.uk/shoptalk/how_to_compete_so_youll_always_win/you_cant_win_a_fight
Forest
https://www.flickr.com/photos/nikola-ostrun/14827652585
Engine
http://www.classictrucks.com/tech/1310_chevy_small_block_engine_bottom_end/
Feature engineering
http://www.globalpost.com/dispatch/news/regions/americas/united-states/131227/the-top-10-most-read-globalpost-stories-2013
Methodology
https://www.flickr.com/photos/omarrr/8450061945
Taxis
https://www.flickr.com/photos/shedboy/3681317392
http://forcailini.blogspot.com.es/2013_06_01_archive.html
Iris
http://en.wikipedia.org/wiki/Iris_ensata
Atlas
http://fr.wikipedia.org/wiki/Rockefeller_Center
Crowd
https://www.flickr.com/photos/onesevenone/12731449695
School bus
https://www.flickr.com/photos/digiart2001/4399252227
Joey
http://www.therichest.com/expensive-lifestyle/entertainment/top-10-hottest-girls-that-joey-dated-on-friends/
http://www.razorcinema.com/21-reasons-joey-from-friends-is-awesome/
http://www.youthconnectmag.com/2014/08/27/joey-tribbiani-is-adorable/
Street name signs
Of a friend, Raquel, during our trip to NYC
Barney
http://origin.slapmagazine.com/component/option,com_jfusion/Itemid,4/index.php?topic=64405.0
Path
https://www.flickr.com/photos/shawnhaines/14752153189
Subway
http://stewartmader.com/in-new-york-online-updates-on-your-subway-ride/
Lyft
http://www.businessinsider.com.au/tlc-and-attorney-general-restraining-order-lyft-2014-7
Optimus
http://s439.photobucket.com/user/SAORS/media/optimus.png.html

Transcript

1. a B.Sc. CS thesis by José M. Camacho supervised by Fernando Sancho Ph.D. September 18, 2014 practical approach a MACHINE LEARNING
2. $> whoami José M. Camacho Full-stack developer at the CulturePlex Lab hello@josemazo.com @josemazo josemazo
3. Content Introduction: Why Machine Learning? The study contents BMO: An isolated Machine Learning toolbox Predicting New York City taxi tips Last comments
4. Introduction: Why Machine Learning?
5. Awesome apps
6. Seven bridges of Königsberg
7. That’s it?
8. Intern student Travel salesman problem
9. The study contents
10. Learning through practice
11. A practical resource A theoretical supplement
12. What is learning?
13. What do you learn from?
14. Types of learning
15. Types of problems
16. k-means The algorithms:
17. k-nearest neighbors The algorithms:
18. Linear regression The algorithms:
19. Naive Bayes The algorithms:
20. Decision tree The algorithms:
21. Support vector machine The algorithms:
22. Artificial neural network The algorithms:
23. Ensemble methods The algorithms:
24. Random forest
25. Enough?
26. Measuring the performance
27. Feature engineering
28. Methodology
29. enough? Again,
30. BMO: An isolated Machine Learning toolbox
31. Main requirements: A programming language
32. Main requirements: Easy installation
33. Main requirements: Isolated environment
34. Main requirements: Access from the web
35. Wait!
36. Vs.
37. IPython Notebook example
38. Vs.
39. Main libraries
40. Not bad
41. Open Source
42. Predicting New York City taxi tips
43. Iris, please no…
44. Where is this precious data?
45. Thank you @chris_whong
46. The data: 173,179,759 trips
47. The data: 24 CSV files
48. The data: 48.6 GB
49. A little bit big, isn’t it?
50. Features: Just a lot
51. Features: With a lot of problems
52. Human data: too much noise
53. For example: Negative values
54. For example: Or huge ones
55. For example: Water taxis?
56. For example: August trips
57. For example: Tip > $0
58. TipNotFound exception
59. Solutions: Ordinary values
60. Solutions: Use your experience
61. After the cleaning: 88,156,805 trips
62. After the cleaning: 17.3 GB
63. Let’s learn!
64. Learning: 1,000,000 trips
65. Learning: Regression?
66. Learning: The classes
67. The results: 52.85% of accuracy
68. A disaster you say?
69. Learning again: A look at the tips
71. Learning again: New classes
72. The new results: 71.74% of accuracy
73. Without passenger info?
74. It’s show time!
75. Of course, it’s Open Source
76. Last comments
77. Well done kiddo To myself:
78. Just the beginning
79. by joining forces Improve the taxi dataset
80. MTA
81. Other services
82. Kaggle competitions
83. Improve BMO
84. thanks! A billion