0
1
00:00:01,030 --> 00:00:08,590
Welcome to the first module! In this module we'll learn about how to use data science to solve a practical
1

2
00:00:08,590 --> 00:00:16,840
problem and we will also introduce our first machine learning algorithm - namely linear regression. Regression
2

3
00:00:16,930 --> 00:00:21,920
is the workhorse of data science, business analytics and research.
3

4
00:00:22,180 --> 00:00:28,120
You'll find people using regression whenever someone is looking to model a phenomenon or make a prediction
4

5
00:00:28,390 --> 00:00:31,620
or discover the hidden relationships among things.
5

6
00:00:31,630 --> 00:00:36,410
Also, I want to show you how to solve problems using Data Science.
6

7
00:00:36,430 --> 00:00:38,980
So what kind of problem are you going to look at?
7

8
00:00:38,980 --> 00:00:45,220
Well, imagine that you're a film producer. You provide money to make movies and when the movies make money,
8

9
00:00:45,790 --> 00:00:46,850
you make money.
9

10
00:00:47,140 --> 00:00:50,620
And when the movies lose money, you lose money.
10

11
00:00:50,620 --> 00:00:52,930
But today you discover a problem.
11

12
00:00:53,230 --> 00:00:58,710
You found out that you just put a whole bunch of money into a movie called "Zombies Rise Again
12

13
00:00:58,710 --> 00:00:59,800
Part 5".
13

14
00:01:00,430 --> 00:01:07,030
And if that doesn't sound like a terrible title for a film, then I don't know what does. Worse yet, you've
14

15
00:01:07,030 --> 00:01:09,070
staked your reputation on this film,
15

16
00:01:09,130 --> 00:01:11,990
you've mortgaged your house, you've sold your car,
16

17
00:01:12,340 --> 00:01:15,830
you've sold your grandmother's silverware to fund this film.
17

18
00:01:15,880 --> 00:01:17,740
So, is your money gone?
18

19
00:01:17,740 --> 00:01:19,210
Will you go bankrupt?
19

20
00:01:19,210 --> 00:01:22,490
Well, to answer these questions you would have to be able to predict the future.
20

21
00:01:22,810 --> 00:01:25,710
And this is where data science will come to the rescue.
21

22
00:01:26,410 --> 00:01:33,700
Let's have a think about how a data scientist would approach this problem. Well first, a data scientist
22

23
00:01:33,940 --> 00:01:38,910
would carefully formulate the question that they're looking to answer. Why?
23

24
00:01:39,000 --> 00:01:46,540
Well, a clear and well formulated question will determine the research and it will also affect the kind
24

25
00:01:46,540 --> 00:01:50,900
of data that you will go out and gather. In fact,
25

26
00:01:50,900 --> 00:01:59,930
step 2 is gathering the data that will help us answer the question. But real world data is also messy.
26

27
00:01:59,930 --> 00:02:02,030
So we have to clean the data.
27

28
00:02:02,030 --> 00:02:05,260
We have to look out for missing data or incomplete data.
28

29
00:02:05,270 --> 00:02:08,880
We have to look out for errors and even bound formatting.
29

30
00:02:08,990 --> 00:02:12,530
In fact, we have to explore the data that we've gathered.
30

31
00:02:12,530 --> 00:02:17,540
And often this means visualizing the data so that we can better understand what it is that we're working
31

32
00:02:17,540 --> 00:02:23,860
with. A graph or a chart is much more helpful than a table of numbers.
32

33
00:02:23,960 --> 00:02:25,730
And after we've done all that,
33

34
00:02:25,730 --> 00:02:33,110
the next step is actually training our algorithm using our computer to identify patterns in the data.
34

35
00:02:33,110 --> 00:02:37,000
In our case, that algorithm will be our linear regression.
35

36
00:02:38,090 --> 00:02:41,420
And finally, we have to evaluate the results.
36

37
00:02:41,480 --> 00:02:43,240
How did our algorithm do?
37

38
00:02:43,350 --> 00:02:47,090
Did it answer our question? How accurate was our algorithm
38

39
00:02:47,090 --> 00:02:53,900
in answering our question? The process that I've outlined here is the process data scientists use to
39

40
00:02:53,900 --> 00:02:55,470
solve problems.
40

41
00:02:55,490 --> 00:02:59,850
This is their workflow for understanding and making sense of the world.
41

42
00:02:59,960 --> 00:03:01,800
So let's tackle that first step.
42

43
00:03:01,850 --> 00:03:04,660
Let's formulate our question.
43

44
00:03:04,730 --> 00:03:06,450
So here's our first shot at this.
44

45
00:03:06,500 --> 00:03:09,830
How much money will our movie make? Now,
45

46
00:03:10,040 --> 00:03:12,830
this is actually not a very good question.
46

47
00:03:12,830 --> 00:03:16,260
The problem is is that this question is too vague.
47

48
00:03:16,400 --> 00:03:19,970
For starters, what do we actually mean with the word money?
48

49
00:03:20,060 --> 00:03:21,350
Can we be more specific?
49

50
00:03:22,280 --> 00:03:23,570
How about this?
50

51
00:03:23,570 --> 00:03:27,220
How about "How much revenue will our movie make?"? Now,
51

52
00:03:27,230 --> 00:03:34,040
this is already a much better question, because revenue is something that means something very specific
52

53
00:03:34,250 --> 00:03:35,510
in the business world.
53

54
00:03:35,540 --> 00:03:41,330
It's something that accountants actually measure and track and that means that we can actually find
54

55
00:03:41,330 --> 00:03:43,610
data on movie revenue.
55

56
00:03:43,880 --> 00:03:47,350
But, what does movie revenue actually depend on?
56

57
00:03:47,390 --> 00:03:48,950
Let's have a think about this.
57

58
00:03:49,040 --> 00:03:55,820
Maybe it depends on the actors. Do movies starring Scarlett Johansson do particularly well? Or maybe it depends
58

59
00:03:55,820 --> 00:04:01,470
on the script? Or maybe what actually matters is the film's overall budget?
59

60
00:04:01,790 --> 00:04:07,610
Because after all, you have to pay good actors and you have to pay for special effects and marketing
60

61
00:04:07,610 --> 00:04:09,310
costs a lot of money too.
61

62
00:04:09,350 --> 00:04:14,480
So maybe the overall budget would be the best thing to look at.
62

63
00:04:14,510 --> 00:04:22,040
So if that's the case then we have a hypothesis - the movie budget could indeed be the thing to investigate.
63

64
00:04:23,350 --> 00:04:28,930
After all, there are many famous examples of big budget films that did exceptionally well at the box
64

65
00:04:28,930 --> 00:04:29,790
office.
65

66
00:04:29,950 --> 00:04:38,240
Movies like Avatar, Titanic and The Avengers cost an absolute fortune, but they were enormously successful.
66

67
00:04:38,380 --> 00:04:44,200
But can the success of these movies tell us something about the success of our movie?
67

68
00:04:44,200 --> 00:04:52,090
If we knew how much movies like Avatar or The Avengers spent and how much they made, can we predict how
68

69
00:04:52,090 --> 00:04:53,420
much we'll make
69

70
00:04:53,470 --> 00:04:59,260
based on how much we're spending? So let's go back to our original question and let's make it a little
70

71
00:04:59,260 --> 00:05:00,720
bit more specific.
71

72
00:05:00,760 --> 00:05:06,060
"Can we use movie budgets to predict movie revenue?". Now,
72

73
00:05:06,220 --> 00:05:08,380
that's already a much better question.
73

74
00:05:08,410 --> 00:05:09,280
Why?
74

75
00:05:09,280 --> 00:05:15,910
Well, it's testable, and by testable, I mean that we can actually check if there is a relationship between
75

76
00:05:15,910 --> 00:05:18,760
movie budgets and movie revenue.
76

77
00:05:18,760 --> 00:05:23,630
This is something that we can measure using data and linear regression.
77

78
00:05:23,890 --> 00:05:29,590
And as data scientists we can actually look at this question and identify exactly what it is that we're
78

79
00:05:29,590 --> 00:05:36,930
trying to predict, namely the movie revenue. A data scientist would call this the dependent variable and
79

80
00:05:36,970 --> 00:05:40,000
in machine learning this would be called the target.
80

81
00:05:40,000 --> 00:05:46,930
Also, we can identify what it is that we're using to make the prediction, namely the movie budgets. A data
81

82
00:05:46,930 --> 00:05:53,410
scientist would call the budgets the independent variable and in machine learning this would be called
82

83
00:05:53,500 --> 00:05:54,620
a feature.
83

84
00:05:54,670 --> 00:05:58,600
So now that we've formulated our question we can move on to Step Two.
84

85
00:05:58,750 --> 00:06:05,320
We can go out there and gather some data. And we'll do just that in the next lesson.
85

86
00:06:05,350 --> 00:06:06,120
I'll see you there.