Showing Revision 3 created 05/24/2016 by Udacity Robot.

1. タイトル：
   Aude Explores Coordinated Migration - Data Analysis with R
2. 概説：

3. 670 0:01 - 0:04
   We want you to develop a mindset of being both curious and skeptical, when
4. 4220 0:04 - 0:08
   you work with data. To help you get into this mindset, I want to share
5. 7510 0:08 - 0:11
   another conversation that I had with Aude. In this next video, I want you
6. 11250 0:11 - 0:13
   to listen to Aude's work and look
7. 12860 0:13 - 0:16
   out for how she demonstrated this exact mindset.
8. 15600 0:16 - 0:21
   >> So we gathered all the home towns and current cities from the
9. 20880 0:21 - 0:26
   users and I was looking at conditional probabilities given
10. 26160 0:26 - 0:29
    a hometown. What is the probability that you currently live
11. 29420 0:29 - 0:34
    in each different cities? Like, for example given that your
12. 34330 0:34 - 0:37
    hometown is New York, what is the probability that you
13. 36970 0:37 - 0:39
    live in Chicago or that you live in, that you
14. 39110 0:39 - 0:41
    still live in New York or that you live in
15. 40620 0:41 - 0:43
    San Francisco or Paris and so on. And what I
16. 43490 0:43 - 0:48
    was expecting is that, at least, the most likely city,
17. 47830 0:48 - 0:51
    where you would live right now would be the city where
18. 51180 0:51 - 0:54
    your home town is. If you grew up in
19. 53570 0:54 - 0:56
    Chicago, the most likely place that you're going to
20. 56330 0:56 - 0:59
    be now is still Chicago. You could be moving
21. 59440 0:59 - 1:02
    as well but the most likely place would remain your
22. 62030 1:02 - 1:06
    hometown. But I saw a fair number of cases
23. 65880 1:06 - 1:09
    where the most likely current city was different from
24. 69130 1:09 - 1:12
    the home town and, and that was, was a
25. 72080 1:12 - 1:17
    fairly high probability. I was really surprised. I was wondering
26. 76570 1:17 - 1:22
    if I had, had a prime in my computations, If there was some issues upstream of
27. 81840 1:22 - 1:28
    what I was doing. So I decided to put all the cities on a map. All the pairs of
28. 88280 1:28 - 1:32
    hometowns and current cities for which the most
29. 92400 1:32 - 1:36
    likely current city was different from the hometown. And
30. 95960 1:36 - 1:39
    what we saw on this map was really fascinating
31. 99200 1:39 - 1:42
    because it was really not what we we're expecting.
32. 101760 1:42 - 1:46
    It was not a bug in the code. We were really seeing patterns arise. Here we only
33. 106340 1:46 - 1:49
    plotted pairs of hometown and current city, so
34. 108980 1:49 - 1:52
    there's no movement between the pairs but what we
35. 111760 1:52 - 1:57
    see is that a lot of these cities for which the most likely
36. 116550 1:57 - 2:02
    current city is different from the hometown arise in western Africa or in
37. 121510 2:02 - 2:07
    India or in like Turkey, which we were not
38. 126940 2:07 - 2:11
    necessarily expecting at the beginning. And there were a
39. 130780 2:11 - 2:14
    lot of small cities all moving to the same current
40. 134240 2:14 - 2:17
    city and so we decided to dig a bit
41. 136620 2:17 - 2:20
    more into it. One thing that happens is that sometimes
42. 139590 2:20 - 2:23
    the distribution of the current city is very flat.
43. 143250 2:23 - 2:27
    Given that you grew up in, let's say Paris, maybe
44. 147100 2:27 - 2:29
    you're still living in Paris but maybe you live in
45. 149165 2:29 - 2:32
    like one of the thousand cities in the suburbs and
46. 152420 2:32 - 2:35
    so the distribution is really flat and so we
47. 155230 2:35 - 2:39
    have to decide what was considered as a coordinates
48. 158615 2:39 - 2:42
    demarcation. We decided yeah, the probability to move to
49. 161990 2:42 - 2:45
    that city is high enough that we're considering that.
50. 166180 2:46 - 2:52
    And the other thing we have to think about is that if you look at the map at
51. 171540 2:52 - 2:54
    the world scale or if you zoom to a
52. 174120 2:54 - 2:58
    very specific area, you don't want to see the same things.
53. 177515 2:58 - 3:05
    So,we also want it to have interactivity in the visualization. And so we
54. 184600 3:05 - 3:12
    decided to use D3, which is a Javascript-based visualization framework, which
55. 191650 3:12 - 3:15
    enables you to have a lot of interactivity with with your data
56. 195340 3:15 - 3:19
    and enabled us to do a lot of that exploration and so on.