﻿1
00:00:00,100 --> 00:00:00,900
Hello my friends.

2
00:00:00,900 --> 00:00:05,400
Congratulations again for completing part
one Data preprocessing.

3
00:00:05,500 --> 00:00:08,533
You are now ready to build machine
learning models

4
00:00:08,700 --> 00:00:13,066
and the first models that we're going
to build together are regression models.

5
00:00:13,100 --> 00:00:16,100
Welcome to part two on regression.

6
00:00:16,166 --> 00:00:19,633
This is the branch of machine
learning that aims

7
00:00:19,633 --> 00:00:23,833
to predict some continuous real numbers,
like for example,

8
00:00:23,833 --> 00:00:28,866
a salary or a temperature
or any kind of continuous numerical value.

9
00:00:29,066 --> 00:00:32,933
And together we will build the best models
to make these predictions.

10
00:00:33,366 --> 00:00:37,400
And so now first of all, let's
make sure we are all on the same page.

11
00:00:37,500 --> 00:00:41,033
This is the whole machine learning A
to z folder

12
00:00:41,033 --> 00:00:44,900
containing all the codes in Python
and R and the data set.

13
00:00:45,200 --> 00:00:49,433
I gave you the link of this folder
right before this tutorial.

14
00:00:49,433 --> 00:00:52,500
You know in the article
to make sure to connect to this folder.

15
00:00:52,600 --> 00:00:55,233
And also make sure
to download the whole folder

16
00:00:55,233 --> 00:00:58,933
in order to get the data sets,
because we will have indeed to import them

17
00:00:58,933 --> 00:01:01,300
whenever we build a machine
learning model.

18
00:01:01,300 --> 00:01:02,200
All right.

19
00:01:02,200 --> 00:01:05,533
And so now, since
we're all on the same page, let's start

20
00:01:05,566 --> 00:01:10,133
our new journey within regression part
two regression.

21
00:01:10,133 --> 00:01:11,000
There we go.

22
00:01:11,000 --> 00:01:13,633
There are several sections
inside each section

23
00:01:13,633 --> 00:01:16,633
corresponding to a different regression
model.

24
00:01:16,733 --> 00:01:19,733
We're going to start with simple linear
regression,

25
00:01:19,733 --> 00:01:22,833
which is the simplest machine
learning model you could ever build.

26
00:01:22,833 --> 00:01:25,933
And so it's good that we start with this,
because indeed with this one

27
00:01:25,933 --> 00:01:28,200
we will only have one
independent variable.

28
00:01:28,200 --> 00:01:31,966
You know, one feature and of course one
continuous real value to predict.

29
00:01:32,266 --> 00:01:35,266
Then we will move on to multiple linear
regression,

30
00:01:35,266 --> 00:01:38,733
which is based on the same equation
than simple linear regression.

31
00:01:38,933 --> 00:01:42,200
Only this time we will have
several features instead of one.

32
00:01:42,633 --> 00:01:45,433
Then we will move on
to polynomial regression,

33
00:01:45,433 --> 00:01:49,300
which will allow us to tackle
some non-linear data sets.

34
00:01:49,300 --> 00:01:53,633
You know, data sets with non-linear
correlations as opposed to previous

35
00:01:53,633 --> 00:01:56,633
models, simple linear
regression and multiple linear regression,

36
00:01:56,700 --> 00:02:00,866
which can provide some amazing and
accurate predictions for linear data set.

37
00:02:00,900 --> 00:02:03,600
You know,
data sets with linear correlations.

38
00:02:03,600 --> 00:02:07,266
Then after polynomial regression,
we will move on to support

39
00:02:07,266 --> 00:02:10,900
vector regression,
which is another kind of non-linear model

40
00:02:10,900 --> 00:02:13,100
that can make some accurate predictions

41
00:02:13,100 --> 00:02:16,200
for non-linear
data sets with non-linear correlations.

42
00:02:16,666 --> 00:02:19,400
And then finally we will move on
to decision tree regression

43
00:02:19,400 --> 00:02:22,866
and random forest regression,
which can provide an alternative

44
00:02:23,033 --> 00:02:26,033
to predict an outcome for non-linear
data sets.

45
00:02:26,066 --> 00:02:28,000
So you will have many options.

46
00:02:28,000 --> 00:02:29,633
You know, after this part two,

47
00:02:29,633 --> 00:02:33,666
you will have basically a toolkit
of several regression models.

48
00:02:33,666 --> 00:02:36,966
And so whenever you end up with a new data
set where you have to predict

49
00:02:36,966 --> 00:02:40,366
a real continuous outcome, well,
you can just try all of them

50
00:02:40,366 --> 00:02:43,933
and select in the end
the one that has the best accuracy.

51
00:02:44,166 --> 00:02:47,400
And thanks to the code templates
that will result

52
00:02:47,400 --> 00:02:50,633
from each of these sections,
you know you will get very clear code

53
00:02:50,633 --> 00:02:54,000
templates that you can adapt
to your own data sets so that you can try

54
00:02:54,000 --> 00:02:57,866
these models on your own data sets
in a flashlight quickly and efficiently,

55
00:02:57,933 --> 00:03:01,666
so that you can select the best one,
giving you the best accuracy.

56
00:03:02,100 --> 00:03:03,666
And so now in this first section

57
00:03:03,666 --> 00:03:07,233
in regression, we're going to start
with simple linear regression of course.

58
00:03:07,433 --> 00:03:08,266
So there we go.

59
00:03:08,266 --> 00:03:10,266
Make sure to go inside this folder.

60
00:03:10,266 --> 00:03:11,700
We're going to start with Python.

61
00:03:11,700 --> 00:03:16,266
So let's go inside this Python folder
and you will find inside two files.

62
00:03:16,466 --> 00:03:19,900
First the data set salary data dot CSV.

63
00:03:20,200 --> 00:03:24,800
And of course our Python implementation
which has the IPython notebook format

64
00:03:24,800 --> 00:03:29,100
which I remind you can run on either
Google Colab or Jupyter Notebook.

65
00:03:29,400 --> 00:03:33,433
And of course we're going to implement
this model together on Google Colab.

66
00:03:33,433 --> 00:03:36,566
But first let me explain the data set
and the problem

67
00:03:36,566 --> 00:03:39,566
that we're going to solve
with simple linear regression.

68
00:03:39,566 --> 00:03:42,600
And to do this, I'm going to open our data

69
00:03:42,600 --> 00:03:45,600
set here and explain what this is about.

70
00:03:45,700 --> 00:03:46,200
All right.

71
00:03:46,200 --> 00:03:50,800
So first I want to reassure you that
indeed this is a very simple data set.

72
00:03:50,800 --> 00:03:52,900
You know, with only 30 observations.

73
00:03:52,900 --> 00:03:55,800
And of course in real life
the data sets are more complex.

74
00:03:55,800 --> 00:03:58,766
But I want to start
working on a simple data set

75
00:03:58,766 --> 00:04:02,300
so that we can really focus on
how to build the model itself.

76
00:04:02,333 --> 00:04:04,500
You know,
because if we had a complex data set,

77
00:04:04,500 --> 00:04:07,233
we would lose a bit
our focus on the model.

78
00:04:07,233 --> 00:04:09,500
And I really want us
to focus on the model.

79
00:04:09,500 --> 00:04:11,633
So let me describe the data set.

80
00:04:11,633 --> 00:04:13,600
It is a data set containing,
as you can see,

81
00:04:13,600 --> 00:04:17,733
30 observations and two columns with
of course one feature.

82
00:04:17,733 --> 00:04:21,566
This is the feature years of experience
and the dependent variable

83
00:04:21,566 --> 00:04:24,400
which we want to predict
which is the salary.

84
00:04:24,400 --> 00:04:28,900
So let's say that this data set belongs
to a company that gathered

85
00:04:28,900 --> 00:04:32,700
data of some of their employees,
collecting for each of them,

86
00:04:32,833 --> 00:04:36,000
their years of experience
and their salary.

87
00:04:36,000 --> 00:04:41,700
So you see, each row of the data set here
corresponds to different employees,

88
00:04:41,700 --> 00:04:43,900
corresponds
to one employee of the company,

89
00:04:43,900 --> 00:04:47,766
and for each employee of this company,
we have indeed the number of years

90
00:04:47,766 --> 00:04:50,833
of experience in the company
and their salary.

91
00:04:50,833 --> 00:04:51,733
Okay.

92
00:04:51,733 --> 00:04:56,333
And so the goal very simply is to build
a simple linear regression model

93
00:04:56,600 --> 00:05:00,233
that will be trained to understand
the correlations

94
00:05:00,233 --> 00:05:03,933
between the number of years of experience
and the salary,

95
00:05:04,166 --> 00:05:08,866
so that it can predict for a new employee,
you know, having a new number

96
00:05:08,866 --> 00:05:10,033
of years of experience.

97
00:05:10,033 --> 00:05:14,066
Well, the corresponding salary
or the salary that this person should get.

98
00:05:14,366 --> 00:05:16,233
So you see, that's a very easy problem.

99
00:05:16,233 --> 00:05:19,833
But at least you'll know how to build
a simple linear

100
00:05:19,833 --> 00:05:23,100
regression model
and perfectly master it in all the detail.

101
00:05:23,466 --> 00:05:25,766
And then I want to say something
very important.

102
00:05:25,766 --> 00:05:28,133
Remember
that we're going to build each time

103
00:05:28,133 --> 00:05:31,433
some code templates
which you can adapt to your own data set,

104
00:05:31,766 --> 00:05:34,833
so that when you want to use the code
template on your own data set,

105
00:05:35,000 --> 00:05:39,433
you will only have one thing to change,
which will be the name of your data.

106
00:05:39,666 --> 00:05:44,300
I will always make sure to make the code
templates as much generic as we can,

107
00:05:44,300 --> 00:05:48,066
so that you only have 1 or 2 things
to change when you want to deploy them

108
00:05:48,066 --> 00:05:48,833
on your data set.
