1
00:00:00,066 --> 00:00:02,500
Hello and welcome to this art tutorial.

2
00:00:02,500 --> 00:00:05,666
So now that we fitted our logistic
regression to the training set,

3
00:00:05,966 --> 00:00:09,533
we are going to predict
the test set results using this classifier

4
00:00:09,533 --> 00:00:12,266
that we just built
in the previous tutorial.

5
00:00:12,266 --> 00:00:14,766
So usually we can do this in one step.

6
00:00:14,766 --> 00:00:17,700
But for logistic regression
we're going to do it in two steps.

7
00:00:17,700 --> 00:00:21,500
And we are more interested
in having the zero and one predictions

8
00:00:21,833 --> 00:00:23,633
rather than the probabilities.

9
00:00:23,633 --> 00:00:25,733
So first let's take anyway
the probabilities.

10
00:00:25,733 --> 00:00:28,733
So we are going to call them prob

11
00:00:29,100 --> 00:00:31,966
pred which is the vector of the,

12
00:00:31,966 --> 00:00:36,166
which will be the vector of the predicted
probabilities of our test set.

13
00:00:36,333 --> 00:00:38,500
Observations by the classifier.

14
00:00:38,500 --> 00:00:41,500
So prop read equals predict.

15
00:00:41,766 --> 00:00:45,300
So we're going to use this predict
function to predict the probabilities

16
00:00:45,566 --> 00:00:48,566
based on our GLM classifier.

17
00:00:48,700 --> 00:00:52,833
And speaking of classifier this is our
first argument in the predict function.

18
00:00:52,833 --> 00:00:55,066
So here we input classifier

19
00:00:56,500 --> 00:00:57,000
come up.

20
00:00:57,000 --> 00:01:00,000
Then the next argument is type.

21
00:01:00,933 --> 00:01:05,366
And for logistic regression
we should choose the response type

22
00:01:05,800 --> 00:01:09,166
because that will give us the
probabilities listed in a single vector.

23
00:01:09,666 --> 00:01:11,900
So that's what we want.

24
00:01:11,900 --> 00:01:16,100
And then we have to specify the new
observations that we want to predict.

25
00:01:16,633 --> 00:01:19,200
So in our case these new observations

26
00:01:19,200 --> 00:01:22,200
are the test set observations.

27
00:01:22,200 --> 00:01:25,366
So new data equals test set.

28
00:01:26,866 --> 00:01:29,866
And we're just going to remove
the last column

29
00:01:29,900 --> 00:01:33,966
of the test set because the last column
is the dependent variable.

30
00:01:33,966 --> 00:01:35,900
And that's what we want to predict.

31
00:01:35,900 --> 00:01:38,100
So we will only take test set minus three.

32
00:01:38,100 --> 00:01:42,033
That is the two independent variables age
and salary of the test set.

33
00:01:42,800 --> 00:01:45,333
All right
so let's just summarize this line.

34
00:01:45,333 --> 00:01:48,733
We are predicting
the test set observations

35
00:01:49,133 --> 00:01:53,100
using our classifier which is the logistic
regression classifier.

36
00:01:54,266 --> 00:01:54,600
All right.

37
00:01:54,600 --> 00:01:58,566
But that returns the probabilities because
as you can see if I select this line

38
00:01:58,566 --> 00:02:01,566
and execute here if I type prob

39
00:02:02,400 --> 00:02:05,400
pred in the console I will get.

40
00:02:06,233 --> 00:02:09,300
As you can see, I will get
the probabilities of each of my test

41
00:02:09,300 --> 00:02:11,033
set observations.

42
00:02:11,033 --> 00:02:14,033
So here, for example,
that's let's go to our test set.

43
00:02:14,533 --> 00:02:17,533
The first observation has index two.

44
00:02:18,066 --> 00:02:19,833
The real result is zero.

45
00:02:19,833 --> 00:02:23,533
That means that the user number two
didn't buy the SUV.

46
00:02:23,966 --> 00:02:26,833
And now
if we go back to logistic regression

47
00:02:26,833 --> 00:02:29,866
we can see that for this user number two

48
00:02:30,033 --> 00:02:33,033
prob pred returns 0.0 16.

49
00:02:33,100 --> 00:02:34,900
So what is this probability exactly.

50
00:02:34,900 --> 00:02:39,600
This probability is the probability that
the dependent variable is equal to one.

51
00:02:39,900 --> 00:02:43,866
That is the probability that the user buys
the SUV.

52
00:02:44,266 --> 00:02:47,266
So here, since it's very small,
it's 0.0 16.

53
00:02:47,600 --> 00:02:50,566
That means that the classifier predicts
a very low

54
00:02:50,566 --> 00:02:54,000
probability of the dependent variable
being equal to one.

55
00:02:54,266 --> 00:02:57,066
That means that it predicts
that the user number two

56
00:02:57,066 --> 00:02:59,833
has very low chance to buy the SUV.

57
00:03:00,966 --> 00:03:03,666
Therefore, in short, prop pred returns

58
00:03:03,666 --> 00:03:08,400
the predicted probabilities
that the user will buy the SUV.

59
00:03:09,033 --> 00:03:10,400
But we don't really want that.

60
00:03:10,400 --> 00:03:13,400
We would rather
have the zero and one result.

61
00:03:13,466 --> 00:03:14,933
And to do this, it's very simple.

62
00:03:14,933 --> 00:03:17,933
We're just going to do some kind of a
conversion.

63
00:03:19,033 --> 00:03:21,900
So we're going to create a vector
of predicted results.

64
00:03:21,900 --> 00:03:24,900
That's we're going to call Y pred

65
00:03:26,066 --> 00:03:27,000
equals.

66
00:03:27,000 --> 00:03:30,300
And then we're just going to use ifelse
to transform

67
00:03:30,300 --> 00:03:33,533
those probabilities into zero
and one results.

68
00:03:33,800 --> 00:03:36,800
So here I'm going to write if else.

69
00:03:36,900 --> 00:03:39,900
So the first argument of
if else is the condition.

70
00:03:40,300 --> 00:03:43,300
And the condition is going to be from

71
00:03:43,833 --> 00:03:47,066
pred over 0.5,

72
00:03:48,766 --> 00:03:51,666
because if prob pred is larger than 0.5,

73
00:03:51,666 --> 00:03:56,100
that means that the user has more chance
to buy the SUV.

74
00:03:56,700 --> 00:03:59,700
So in that case,
that means that we want to predict one.

75
00:03:59,766 --> 00:04:03,466
And actually the second argument of the
if else function is the result

76
00:04:03,466 --> 00:04:06,433
you want to take
when this condition is true.

77
00:04:06,433 --> 00:04:08,100
So here we're going to put a one.

78
00:04:09,066 --> 00:04:09,900
All right.

79
00:04:09,900 --> 00:04:13,133
And the last argument is the result
we want to return

80
00:04:13,166 --> 00:04:14,866
if this condition is false.

81
00:04:14,866 --> 00:04:16,133
And if this condition is false

82
00:04:16,133 --> 00:04:20,333
that means that the predicted
probability is lower than 0.5.

83
00:04:20,533 --> 00:04:23,833
That means that the user has less chance
to buy the SUV,

84
00:04:24,266 --> 00:04:27,266
and therefore it's going to be a zero.

85
00:04:27,400 --> 00:04:28,233
And that's it.

86
00:04:28,233 --> 00:04:32,333
Now you're going to see
if I select this line and execute.

87
00:04:33,266 --> 00:04:34,233
All right.

88
00:04:34,233 --> 00:04:37,100
Let's look at wipe Red. Now

89
00:04:37,100 --> 00:04:39,933
wipe red as you can see

90
00:04:39,933 --> 00:04:42,066
I only have zeros or one.

91
00:04:42,066 --> 00:04:44,900
I no longer have the probabilities.

92
00:04:44,900 --> 00:04:48,866
And remember the real result for our user
two was zero.

93
00:04:48,866 --> 00:04:51,866
So that's what happened in reality.

94
00:04:52,900 --> 00:04:55,566
And here our classifier predicted zero.

95
00:04:55,566 --> 00:04:57,633
So that's a correct prediction.

96
00:04:57,633 --> 00:05:01,400
And that's because remember
before the probability was very low.

97
00:05:01,600 --> 00:05:03,566
So it was below 0.5.

98
00:05:03,566 --> 00:05:05,566
And therefore it returned zero.

99
00:05:06,800 --> 00:05:07,200
All right.

100
00:05:07,200 --> 00:05:09,566
So we predicted our test results.

101
00:05:09,566 --> 00:05:10,233
That's a good thing.

102
00:05:10,233 --> 00:05:14,733
Now we are going to evaluate this
prediction.

103
00:05:15,000 --> 00:05:19,100
Thanks to the confusion matrix that
we going to build in the next tutorial.

104
00:05:19,400 --> 00:05:20,866
So I look forward to seeing you there.

105
00:05:20,866 --> 00:05:23,866
And until then enjoy machine learning.