Scorpion Explaining Away Outliers in Aggregate Queries

1. ScorpionExplaining Away Outliers in Aggregate Queries eugenewuand sam madden MIT http://springfieldpunx.blogspot.com/2010/11/mortal-kombat-ninjas-scorpion.html

8. Table Split Visualize Aggregate

9. SELECT sum(cost) FROM expenses GROUPBY country Expenses USA China Italy

10. Expenses USA China Italy SELECT sum(cost) FROM expenses GROUPBY country

11. Expenses USA China Italy SELECT sum(cost) FROM expenses GROUPBY country

12. Expenses USA China Italy SELECT sum(cost) FROM expenses GROUPBY country

13. Given Outlier and normal results Understand Why Expenses USA China Italy SELECT sum(cost) FROM expenses GROUPBY country

14. Given Outlier and normal results What input properties caused the outliers? most caused the outliers? caused outliers but didn’t affect normal outputs? Expenses USA China Italy SELECT sum(cost) FROM expenses GROUPBY country

15. Can’t Touch This

17. Provenance Data!

18. Provenance SELECT SUM(cost) FROM sam’s bank account $$$

19. SELECT SUM(cost) FROM sam’s bank account Provenance $$$

20. Provenance SELECT SUM(cost) FROM sam’s bank account $$$

21. Provenance SELECT SUM(cost) FROM sam’s bank account $$$ Darn! Ya caught me

22. Provenance http://weknowmemes.com/2012/04/whats-the-point/

23. SELECT SUM(cost) FROM sam’s bank account Provenance Filter for “most influential”

24. Provenance

25. Provenance Faceting http://www.perceptualedge.com/articles/Whitepapers/Three_Blind_Men.pdf

26. Provenance Faceting http://www.perceptualedge.com/articles/Whitepapers/Three_Blind_Men.pdf

27. Provenance Faceting Dimensionality :( Dealing with multiple outliers? http://www.perceptualedge.com/articles/Whitepapers/Three_Blind_Men.pdf

28. Provenance Faceting

29. Provenance Faceting Scorpion!

30. Given Outlier and normal results Understand Why Expenses USA China Italy

31. Given Outlier and normal results Find Predicates correlated with outliers Expenses USA China Italy Desc = “toilets”

32. Given Outlier and normal results Removing predicate from inputs “fixes” outliers & maintains normal results Find Predicates correlated with outliers Expenses s.t. USA China Italy Desc = “toilets”

33. Given Outlier and normal results Find Predicates correlated with outliers Expenses s.t. Removing predicate from inputs “fixes” outliers & maintains normal results USA China Italy Desc = “toilets”

34. Given Outlier and normal results Removing predicate from inputs “fixes” outliers & maintains normal results Find Predicates correlated with outliers Expenses s.t. USA China Italy Desc = “toilets”

35. Formalize “influence” as metric Predicate search heuristics Some results

36. T

37. T Desc = “toilet” p(T)

38. T p(T)

39. T p(T) T – p(T)

40. p(T)

41. p(T)

42. Δoutput p(T)

43. Δoutput p(T) |p(T)|

44. Δoutput p(T) |p(T)| Δoutput Influence Metric |p(T)|

45. Δf(x) Sensitivity Analysis Δx Δoutput Influence Metric |p(T)|

46. Δoutput |p(T)| ΔOutput “High vs Low” |p(T)| ΔNormal Multiple Outputs

47. Δoutput |p(T)| ΔOutput “High vs Low” |p(T)| ΔNormal Multiple Outputs ΔoutputV |p(T)|

48. Δoutput |p(T)| ΔOutput “High vs Low” |p(T)| ΔNormal Multiple Outputs ΔoutputV |p(T)| ΔoutputV |p(T)|c

49. Δoutput |p(T)| ΔOutput “High vs Low” |p(T)| ΔNormal Multiple Outputs ΔoutputV |p(T)| ΔoutputV |p(T)|c - ΔoutlierV ΔNormal |p(T)|c

50. Δoutput |p(T)| ΔOutput “High vs Low” |p(T)| ΔNormal Multiple Outputs ΔoutputV |p(T)| ΔoutputV |p(T)|c - ΔoutlierV ΔNormal |p(T)|c - ΔoutlierV max mean ΔNormal |p(T)|c normal outlier