﻿1
00:00:00,600 --> 00:00:01,133
All right.

2
00:00:01,133 --> 00:00:03,300
So that's the results of the training set.

3
00:00:03,300 --> 00:00:04,700
These are quite good results.

4
00:00:04,700 --> 00:00:09,033
But that's because we trained our linear
regression models on the training set.

5
00:00:09,533 --> 00:00:13,466
And now let's see how our simple linear
regression model can predict

6
00:00:13,566 --> 00:00:14,600
new observations.

7
00:00:14,600 --> 00:00:17,400
That is new observations of the test set.

8
00:00:17,400 --> 00:00:19,700
So I'm going to close that okay.

9
00:00:19,700 --> 00:00:23,533
And now let's do the same for the test
sets to see if our simple linear

10
00:00:23,533 --> 00:00:27,300
regression can make some good predictions
on new observations.

11
00:00:27,300 --> 00:00:30,466
So the test set observations
on which our model wasn't built.

12
00:00:30,900 --> 00:00:33,833
So as usual we want to be efficient.

13
00:00:33,833 --> 00:00:36,833
And so we're going to copy all this

14
00:00:38,100 --> 00:00:41,100
copy and paste it here.

15
00:00:41,400 --> 00:00:42,333
Okay.

16
00:00:42,333 --> 00:00:45,333
So let's replace all the training set here

17
00:00:45,733 --> 00:00:48,666
by to set okay

18
00:00:48,666 --> 00:00:49,400
here as well.

19
00:00:49,400 --> 00:00:51,466
Training set to set.

20
00:00:51,466 --> 00:00:53,800
That's the test results okay.

21
00:00:53,800 --> 00:00:56,100
So now we must be careful with something.

22
00:00:56,100 --> 00:00:59,100
Let's see if we have to replace
all the training set.

23
00:00:59,466 --> 00:01:03,500
So in the first component of ggplot
that is when we plot

24
00:01:03,500 --> 00:01:04,700
all the observation points.

25
00:01:04,700 --> 00:01:06,766
The real observations

26
00:01:06,766 --> 00:01:10,366
here we have training set years
experience and training set salary.

27
00:01:10,566 --> 00:01:13,600
In your opinion do we need to change
training set here by test set?

28
00:01:14,166 --> 00:01:15,100
Yes of course.

29
00:01:15,100 --> 00:01:19,566
Because this time we want to have
the observation points of the test set.

30
00:01:20,533 --> 00:01:21,800
So we want to

31
00:01:21,800 --> 00:01:24,800
take the years experience
here of the test set observations

32
00:01:24,933 --> 00:01:27,733
and the salaries
of the test set observations.

33
00:01:27,733 --> 00:01:31,366
And here we will replace training
set by test set okay.

34
00:01:31,366 --> 00:01:33,300
But now in geom line.

35
00:01:33,300 --> 00:01:36,300
So that is the component that builds
a regression line.

36
00:01:36,566 --> 00:01:40,166
Do we need to replace training set by test
set here.

37
00:01:40,333 --> 00:01:43,000
You know training
set here and training set here.

38
00:01:43,000 --> 00:01:46,000
Do we need to replace those two
training sets by test set.

39
00:01:46,700 --> 00:01:49,033
Well the answer is actually of course no

40
00:01:49,033 --> 00:01:52,566
because our regressor
is already trained on the training set.

41
00:01:52,566 --> 00:01:56,333
So whether we keep here
the training set or replaced by test set,

42
00:01:56,566 --> 00:01:59,366
we will obtain the same simple linear
regression line.

43
00:01:59,366 --> 00:02:02,200
Indeed, if we replaced here training
set by test set,

44
00:02:02,200 --> 00:02:05,500
we would just build some new points
of this same regression line

45
00:02:05,500 --> 00:02:08,566
corresponding to the new predictions
of the test set observation points.

46
00:02:08,566 --> 00:02:12,666
Because when we trained our simple linear
regressor on the training set,

47
00:02:13,000 --> 00:02:17,466
we obtained one unique model equation,
which is the simple linear equation

48
00:02:17,466 --> 00:02:18,266
itself.

49
00:02:18,266 --> 00:02:18,866
And therefore,

50
00:02:18,866 --> 00:02:22,800
whether we build the regression line here
by predicting the training set points

51
00:02:22,833 --> 00:02:24,300
or the test set points.

52
00:02:24,300 --> 00:02:27,366
Well, since this predictions result
from the same unique

53
00:02:27,366 --> 00:02:30,533
simple linear equation,
we will get the same regression line.

54
00:02:31,266 --> 00:02:32,166
Okay, now.

55
00:02:32,166 --> 00:02:33,733
So that means that now we're ready okay.

56
00:02:33,733 --> 00:02:37,166
To set to set test
it here and training set here.

57
00:02:37,166 --> 00:02:37,900
We're all good.

58
00:02:37,900 --> 00:02:41,866
We are ready to plot the test set results

59
00:02:42,000 --> 00:02:45,000
to find out how our simple linear

60
00:02:45,000 --> 00:02:48,000
regression behaves on new observations.

61
00:02:48,366 --> 00:02:50,066
So let's do this.
Press Command and control.

62
00:02:50,066 --> 00:02:51,200
Plus enter to execute.

63
00:02:52,866 --> 00:02:54,066
And here we go.

64
00:02:54,066 --> 00:02:58,466
So as you can see as I was telling you
if we go back to the training set here,

65
00:02:58,866 --> 00:03:02,633
you can see that
only the red points change you see.

66
00:03:02,966 --> 00:03:05,100
And the blue regression
line doesn't change.

67
00:03:05,100 --> 00:03:06,666
You see, it's the same here.

68
00:03:06,666 --> 00:03:08,366
Only the red points change.

69
00:03:08,366 --> 00:03:10,800
So here
that's the training set observation points

70
00:03:10,800 --> 00:03:12,766
with the model built on the training set.

71
00:03:12,766 --> 00:03:16,866
And here that's the test set observations
with the same regression line

72
00:03:17,100 --> 00:03:19,000
builds on the training set okay.

73
00:03:19,000 --> 00:03:22,266
So now let's zoom on the test set plots
and give some interpretation

74
00:03:23,766 --> 00:03:24,100
okay.

75
00:03:24,100 --> 00:03:26,233
So the test set results
are actually not bad.

76
00:03:26,233 --> 00:03:29,366
We can see that
we have some very good predictions here.

77
00:03:29,666 --> 00:03:32,666
For example for this employee
this employee this employee

78
00:03:33,300 --> 00:03:34,866
and this one as well.

79
00:03:34,866 --> 00:03:38,800
Because as you can see
the real value is closed to the prediction

80
00:03:38,800 --> 00:03:40,700
which is on the blue regression line.

81
00:03:40,700 --> 00:03:45,433
So since these are new observations
on which our simple linear regression

82
00:03:45,433 --> 00:03:49,133
model didn't learn any correlation,
that's actually quite a good job

83
00:03:49,133 --> 00:03:52,433
to find, you know, predictions
so close to the real value.

84
00:03:52,600 --> 00:03:56,466
And of course, there are some predictions
that are further from the real value.

85
00:03:56,766 --> 00:04:00,300
But that's because there is not, 100%

86
00:04:00,300 --> 00:04:04,200
linear dependency between the salary
and the years of experience.

87
00:04:04,566 --> 00:04:07,566
But for sure there is a certain linear
dependency.

88
00:04:07,966 --> 00:04:08,366
All right.

89
00:04:08,366 --> 00:04:11,466
So thank you for watching this tutorial
about simple linear regression.

90
00:04:11,800 --> 00:04:15,833
In the next section, we will find out
about a new type of linear regression,

91
00:04:16,100 --> 00:04:20,000
which is going to be the multiple linear
regression, meaning that instead

92
00:04:20,000 --> 00:04:24,733
of having only one independent variable,
we'll have several independent variables

93
00:04:25,033 --> 00:04:28,033
that will predict
a single dependent variable.

94
00:04:28,133 --> 00:04:30,933
So there will be a new business problem,
and I can't wait

95
00:04:30,933 --> 00:04:33,933
to solve it with you
using multiple linear regression.

96
00:04:34,300 --> 00:04:35,700
I'll see you in the next section.

97
00:04:35,700 --> 00:04:37,600
And until then, enjoy machine learning.
