1
00:00:00,166 --> 00:00:01,000
Hello my friends.

2
00:00:01,000 --> 00:00:03,400
Welcome to this new practical activity

3
00:00:03,400 --> 00:00:08,033
and the first practical activity
actually of part three classification.

4
00:00:08,266 --> 00:00:11,266
I'm super excited as you might hear
by the sound of my voice,

5
00:00:11,266 --> 00:00:13,633
because this is one of my favorite parts.

6
00:00:13,633 --> 00:00:15,000
And actually you will see that

7
00:00:15,000 --> 00:00:19,466
the case study we will do together
is pretty fun, so I can't wait to start.

8
00:00:19,466 --> 00:00:20,400
Welcome back.

9
00:00:20,400 --> 00:00:21,600
And so now here we go.

10
00:00:21,600 --> 00:00:25,400
We are about to enter a new branch
of machine learning where

11
00:00:25,433 --> 00:00:29,166
this time we want predict
a continuous numerical value

12
00:00:29,166 --> 00:00:30,933
like in part two regression.

13
00:00:30,933 --> 00:00:33,666
But this time we will predict a category.

14
00:00:33,666 --> 00:00:38,866
You know, a class like for example,
you know, a binary variable 0 or 1, one.

15
00:00:38,866 --> 00:00:41,933
The classic example is actually
to do classification to predict

16
00:00:42,033 --> 00:00:45,466
a tumor, you know,
whether a tumor is benign or malignant.

17
00:00:45,466 --> 00:00:48,666
And this is actually the case
that we will do at the end, you know,

18
00:00:48,666 --> 00:00:52,700
when I will teach you on how to select
the best classification model,

19
00:00:52,866 --> 00:00:54,333
but that will come at the end.

20
00:00:54,333 --> 00:00:56,400
First, we will do a fun case study

21
00:00:56,400 --> 00:01:00,233
with which we will learn how to build each
of our classification models.

22
00:01:00,466 --> 00:01:01,800
And there you go my friends.

23
00:01:01,800 --> 00:01:06,000
It is in this tutorial that I will explain
the problem of this case study.

24
00:01:06,666 --> 00:01:09,866
All right, so before we start,
let's just make sure once again

25
00:01:09,866 --> 00:01:12,100
everyone here is on the same page.

26
00:01:12,100 --> 00:01:16,100
So right before this tutorial
I give you the link to this folder

27
00:01:16,100 --> 00:01:19,133
containing all the codes
and data sets in the ten parts.

28
00:01:19,266 --> 00:01:21,533
So make sure to connect to that link.

29
00:01:21,533 --> 00:01:24,300
And now
now we should all be on the same page.

30
00:01:24,300 --> 00:01:25,200
And there you go.

31
00:01:25,200 --> 00:01:28,200
We're going to go into part
three classification

32
00:01:28,300 --> 00:01:32,966
to tackle our first model
logistic regression.

33
00:01:32,966 --> 00:01:37,533
I hope you enjoyed the intuition lectures
and mostly that you are now ready

34
00:01:37,666 --> 00:01:40,966
to put your
well learned theory into practice.

35
00:01:40,966 --> 00:01:45,166
And we're going to put it into practice
first with Python, with which

36
00:01:45,166 --> 00:01:49,300
we're going to re-implement from scratch
and step by step

37
00:01:49,300 --> 00:01:52,400
the whole logistic regression
implementation.

38
00:01:52,800 --> 00:01:55,800
So as you can see in this Python folder,
you have two files.

39
00:01:55,900 --> 00:02:00,633
First, well that logistic regression
implementation in ipynb format

40
00:02:00,633 --> 00:02:04,600
which you can open with either
Google Colaboratory or Jupyter Notebook,

41
00:02:04,966 --> 00:02:10,166
and you have the data set
which is called social network ads.

42
00:02:10,366 --> 00:02:11,533
So let's open it.

43
00:02:11,533 --> 00:02:16,100
And now let me explain
what the problem is about.

44
00:02:16,800 --> 00:02:17,333
All right.

45
00:02:17,333 --> 00:02:20,500
So let's imagine our favorite car company.

46
00:02:20,500 --> 00:02:24,266
I won't mention name here because I
don't want to do any kind of advertising.

47
00:02:24,433 --> 00:02:27,633
But let's imagine
your favorite car company.

48
00:02:27,800 --> 00:02:31,600
And let's imagine that you are
a data scientist for that company.

49
00:02:32,133 --> 00:02:36,066
And your mission, should you choose
to accept it, is to predict

50
00:02:36,166 --> 00:02:39,700
which of your previous customers will buy

51
00:02:39,933 --> 00:02:42,933
a brand new, beautiful SUV

52
00:02:42,933 --> 00:02:46,266
just created by your favorite car company.

53
00:02:46,700 --> 00:02:49,600
All right, so your favorite car company
has just released

54
00:02:49,600 --> 00:02:53,000
this brand new, beautiful,
irresistible SUV.

55
00:02:53,333 --> 00:02:56,766
And the general manager of this car
company has asked you,

56
00:02:56,800 --> 00:02:59,733
you know, the most talented
data scientists of the company

57
00:02:59,733 --> 00:03:03,600
to predict
which customers will buy that new SUV.

58
00:03:03,600 --> 00:03:06,133
You know,
with the highest conversion rate.

59
00:03:06,133 --> 00:03:07,800
And to help you, because, you know,

60
00:03:07,800 --> 00:03:11,566
this general manager has some minimum
data science skills and knows

61
00:03:11,733 --> 00:03:15,700
and understands that in order to predict
this, you need data right?

62
00:03:15,700 --> 00:03:19,500
Your data on which to train
your classification model

63
00:03:19,666 --> 00:03:22,566
to predict what needs to be predicted,
meaning

64
00:03:22,566 --> 00:03:25,566
which customers
will buy that brand new SUV.

65
00:03:25,966 --> 00:03:26,733
And so there you go.

66
00:03:26,733 --> 00:03:30,833
That's exactly the data
that your general manager gave you.

67
00:03:30,966 --> 00:03:32,266
And then this data set.

68
00:03:32,266 --> 00:03:36,600
Well, each
row corresponds to different customers.

69
00:03:36,933 --> 00:03:39,300
And for each of these customers, there
you go.

70
00:03:39,300 --> 00:03:42,200
I'm about to reveal the features
and the dependent variable

71
00:03:42,200 --> 00:03:44,066
for each of these customers.

72
00:03:44,066 --> 00:03:47,066
Well,
this general manager collected the age

73
00:03:47,333 --> 00:03:49,866
and it collected the estimated salary.

74
00:03:49,866 --> 00:03:53,566
Because, you know, when a customer buys
a new car with some kind of credit

75
00:03:53,566 --> 00:03:57,266
or whatever, well, it has to provide
estimated salary in the form.

76
00:03:57,400 --> 00:03:59,800
So that's how it got the estimated salary.

77
00:03:59,800 --> 00:04:02,533
And finally,
that's your dependent variable.

78
00:04:02,533 --> 00:04:06,500
Of course, the purchased variable
telling whether or not

79
00:04:06,666 --> 00:04:09,666
these customers have bought previously

80
00:04:09,700 --> 00:04:13,200
some older
SUVs of this car company. Right.

81
00:04:13,200 --> 00:04:18,200
So this car company has basically many
makes and models of SUVs and all the

82
00:04:18,200 --> 00:04:18,733
zeros.

83
00:04:18,733 --> 00:04:23,066
And the ones that you see here for each
customer is seeing whether or not

84
00:04:23,066 --> 00:04:26,700
these customers
have bought one of these previous SUV

85
00:04:26,933 --> 00:04:30,266
so that your model will be trained
on this data set.

86
00:04:30,500 --> 00:04:34,200
And for new customers,
you know, having a different age

87
00:04:34,200 --> 00:04:36,066
and a different estimated salary.

88
00:04:36,066 --> 00:04:40,633
Well, we will predict
if yes or no, they will buy that new SUV.

89
00:04:41,166 --> 00:04:41,900
Okay.

90
00:04:41,900 --> 00:04:44,900
So of course, in this dependent
variable purchased

91
00:04:45,000 --> 00:04:48,366
zero means that the customer didn't buy
any previous SUV.

92
00:04:48,600 --> 00:04:52,966
And one means that the customer
but some previous SUV's all right.

93
00:04:53,100 --> 00:04:57,466
And therefore all the future predictions
that will be equal to one

94
00:04:57,600 --> 00:05:02,600
will probably mean that the customer
has a high chance to buy the new SUV.

95
00:05:02,800 --> 00:05:04,966
If, of course, we offer a great deal.

96
00:05:04,966 --> 00:05:09,600
And finally, once we predict the customers
that are going to buy the SUV,

97
00:05:09,733 --> 00:05:13,833
well, the final step of the strategy
will be for the advertising team

98
00:05:13,833 --> 00:05:17,700
to post ads of this
brand new SUV on social networks,

99
00:05:17,933 --> 00:05:22,166
and these ads will be targeted
to the customers where we predict one,

100
00:05:22,166 --> 00:05:25,433
you know, where we predict
that they're going to buy that new SUV.

101
00:05:25,600 --> 00:05:25,933
All right.

102
00:05:25,933 --> 00:05:29,866
So you see the idea, the predictive model
will target your customers.

103
00:05:30,033 --> 00:05:33,800
And then the advertising team will use
the results of this predictive model

104
00:05:33,800 --> 00:05:37,333
to optimize the targeting
of future customers.

105
00:05:37,333 --> 00:05:38,366
And that is why,

106
00:05:38,366 --> 00:05:42,733
you know, the name of the data
set is called Social Network AD dot CSV.