Estatística: a gramática de Data Science

Ícaro Medeiros
Especialista em Data Science @Telefônica

Transcript

1. STATISTICS THE GRAMMAR OF DATA SCIENCE ÍCARO MEDEIROS Big Data Week São Paulo - SP, 23/11/2015
2. WHY TO PURSUE SOLID BACKGROUND ON STATISTICS? https://twitter.com/josh_wills/status/198093512149958656
3. INSPIRATIONS FOR THIS KEYNOTE https://speakerdeck.com/jakevdp/statistics-for-hackers https://www.goodreads.com/book/show/17986418-naked-statistics
4. METRICS MEANING METHODS MISUSES - + + +
5. HOW TO LIE: MOVIE REVIEWS http://fivethirtyeight.com/features/fandango-movies-ratings BECAUSE IT LOOKS LIKE MATH, WE [THINK] IT’S SOMEHOW OBJECTIVELY TRUE, BUT IT’S ALL BASED ON SUBJECTIVE EXPERIENCE
6. FANDANGO LOVES MOVIES http://fivethirtyeight.com/features/fandango-movies-ratings (AND SELLS MOVIE TICKETS)
7. ATTENTION TO PROVENANCE
8. HOW TO LIE: ROUNDING http://fivethirtyeight.com/features/fandango-movies-ratings
9. HOW TO LIE: ROUNDING http://fivethirtyeight.com/features/fandango-movies-ratings
10. CORRELATION IS NOT CAUSATION
11. SHORT BREAKS AT WORK
 "CAUSE" CANCER Example from ‘Naked Statistics'
12. SMOKING
 CAUSES CANCER BREAKS AT WORK CORRELATED WITH LEAD TO
13. SPURIOUS CORRELATIONS http://www.tylervigen.com/spurious-correlations
14. A/B TESTING https://vwo.com/ab-testing/
15. https://xkcd.com/882 (Significant) SIGNIFICANCE
18. A/B TESTING CAN BE BAD https://www.quora.com/When-should-A-B-testing-not-be-trusted-to-make-decisions/answer/Edwin-Chen-1 ▸Feedback loops ▸Novelty effect ▸Seasonality ▸Wrong metrics http://www.evanmiller.org/how-not-to-run-an-ab-test.html
19. CHOOSE THE RIGHT METRICS: CLICKS VS DWELL TIME http://yahoolabs.tumblr.com/post/99405569711/science-powering-product-and-personalization
20. CHOOSE THE RIGHT METRICS: SHARING IS NOT NECESSARILY CARING http://time.com/12933/what-you-think-you-know-about-the-web-is-wrong/
21. THE RED CARD PROBLEM http://fivethirtyeight.com/features/science-isnt-broken/ http://www.nature.com/news/crowdsourced-research-many-hands-make-tight-work-1.18508
22. 61 RESEARCHERS: SAME PROBLEM, DIFFERENT METHODS http://fivethirtyeight.com/features/science-isnt-broken/
23. THE BACON CONTROVERSY
24. MORE ABOUT P-VALUES https://twitter.com/Ted_Underwood/status/658983555008040960
25. IS THIS A GOOD CLASSIFICATION? http://www.wired.com/2015/10/who-does-bacon-cause-cancer-sort-of-but-not-really/ 1 CARCINOGENIC 2A PROBABLY 2B POSSIBLY …
26. EFFECT SIZE: BACON VS CIGARETTES (SAME CATEGORY) This is bacon 18% Cigarette
27. WAIT FOR IT…
28. Cigarette 2500% http://www.wired.com/2015/10/who-does-bacon-cause-cancer-sort-of-but-not-really/
29. http://www.theguardian.com/society/2015/oct/26/bacon-ham-sausages-processed-meats-cancer-risk-smoking-says-who https://catracalivre.com.br/geral/sustentavel/indicacao/muito-alem-do-bacon-agrotoxicos-tambem-podem-causar-cancer/
30. THE SCHRODINGER’S DIET http://www.vox.com/2015/5/20/8621527/health-tips-reporter
31. ” http://nerds.airbnb.com/scaling-data-science DATA SCIENCE IS AN ACT OF INTERPRETATION OF CUSTOMER'S VOICE
32. GOOD DATA VISUALIZATION: TIPS FOR SCATTER PLOTS http://content.visage.co/hs-fs/hub/424038/file-2094950163-pdf/Data_Visualization_101_How_to_Design_Charts_and_Graphs.pdf
33. DAVID MCCANDLESS: INFORMATION IS BEAUTIFUL http://www.informationisbeautiful.net/visualizations/diversity-in-tech/
34. IT’S EASY TO LIE WITH STATISTICS, BUT IT’S HARD TO TELL THE TRUTH WITHOUT THEM Andrejs Dunkels, as mentioned on "Naked Statistics" TAKEAWAY MESSAGE
35. slides icaromedeiros.com.br slideshare.net/icaromedeiros @icaromedeiros