1
00:00:00,233 --> 00:00:02,733
Okay,
my friends, are you ready to finish this?

2
00:00:02,733 --> 00:00:06,300
It's actually very simple now,
and I'm sure that you're confident

3
00:00:06,400 --> 00:00:10,400
that the solution I'm about to give
you will be the same as your solution.

4
00:00:10,666 --> 00:00:11,566
Because indeed.

5
00:00:11,566 --> 00:00:15,233
Now, the only thing that we have to do
is juggle with our different

6
00:00:15,300 --> 00:00:18,966
machine learning toolkits,
including the Data preprocessing toolkit

7
00:00:18,966 --> 00:00:23,300
and the classic toolkit
to complete this implementation.

8
00:00:23,333 --> 00:00:24,466
So let's do this.

9
00:00:24,466 --> 00:00:28,200
Starting with splitting the data
set into the training set in a test set.

10
00:00:28,366 --> 00:00:29,800
Well, that's super easy.

11
00:00:29,800 --> 00:00:33,566
We're ready to do this in only one
copy paste because we have indeed

12
00:00:33,566 --> 00:00:36,966
the matrix of features x
and the dependent variable vector y.

13
00:00:37,133 --> 00:00:39,466
Therefore,
the only thing that we have to do here

14
00:00:39,466 --> 00:00:42,700
is just to go to our data
preprocessing template

15
00:00:42,966 --> 00:00:46,366
and then take exactly
these two lines of code

16
00:00:46,633 --> 00:00:52,000
to indeed split our data
set composed of the matrix of features X

17
00:00:52,000 --> 00:00:55,300
and the dependent variable vector y into,
well,

18
00:00:55,300 --> 00:00:58,300
a new training set and test it.

19
00:00:58,366 --> 00:01:00,100
And that's our first copy paste.

20
00:01:00,100 --> 00:01:03,100
And of course here
we have nothing to change.

21
00:01:03,233 --> 00:01:07,133
Now next step training the naive base
model on the training set.

22
00:01:07,333 --> 00:01:10,766
So here we'll have to juggle
with another of our machine

23
00:01:10,766 --> 00:01:13,833
learning toolkit which is of course
the classification toolkit.

24
00:01:14,100 --> 00:01:17,266
So we're going to go back
into our whole machine learning.

25
00:01:17,266 --> 00:01:18,466
Is that folder.

26
00:01:18,466 --> 00:01:19,633
Then we're going to,

27
00:01:19,633 --> 00:01:24,000
you know, use this little shortcut here
to go back to the base of the folder,

28
00:01:24,000 --> 00:01:27,266
which is this one, machine learning A
to Z codes and data sets.

29
00:01:27,600 --> 00:01:30,600
Then we're going to go into part
three classification.

30
00:01:30,600 --> 00:01:35,200
And then we'll see all our different
models including the Naive Bayes.

31
00:01:35,500 --> 00:01:39,300
But I would just like to remind that,
you know, the choice of Naive

32
00:01:39,300 --> 00:01:41,766
Bayes was just based on my experience.

33
00:01:41,766 --> 00:01:45,800
I observed that the Naive Bayes does
very well with natural language

34
00:01:45,800 --> 00:01:47,166
processing problems.

35
00:01:47,166 --> 00:01:50,766
But I'll give you another exercise
at the end of this tutorial,

36
00:01:50,900 --> 00:01:54,266
which will be to beat me or,
you know, to beat the score

37
00:01:54,300 --> 00:01:57,200
that we're going
to get at the end of this implementation.

38
00:01:57,200 --> 00:02:00,600
And so your goal will be
to get an even better accuracy.

39
00:02:00,866 --> 00:02:03,433
And if you get it,
you'll post it in the comments.

40
00:02:03,433 --> 00:02:06,900
Or you can send me a private message
to say that indeed, you managed

41
00:02:06,900 --> 00:02:09,900
to beat the accuracy score
that we're about to get together

42
00:02:09,966 --> 00:02:13,566
and that you probably got by yourself
when doing this exercise.

43
00:02:14,133 --> 00:02:15,366
All right, so there we go.

44
00:02:15,366 --> 00:02:17,700
Let's just choose for now
the Naive Bayes model.

45
00:02:17,700 --> 00:02:20,933
So we're going to go into the section
section 18 Naive Bayes.

46
00:02:21,266 --> 00:02:23,100
Then we're going to go into Python.

47
00:02:23,100 --> 00:02:26,533
And then we're going to open this
Naive Bayes implementation

48
00:02:26,833 --> 00:02:29,166
open with Google Collaboratory.

49
00:02:29,166 --> 00:02:32,166
And you know I can just put it here.

50
00:02:32,466 --> 00:02:35,866
It is now opening
loading it laying out the notebook.

51
00:02:36,100 --> 00:02:36,933
And there we go.

52
00:02:36,933 --> 00:02:38,866
And now you have everything.

53
00:02:38,866 --> 00:02:42,933
You can just find the cell
that's, you know, trained Naive Bayes

54
00:02:42,933 --> 00:02:47,100
model on the training set,
which is right here.

55
00:02:47,100 --> 00:02:50,133
By the way, you could also take,
you know, your model selection folder

56
00:02:50,133 --> 00:02:53,133
containing
all the classification models as you want.

57
00:02:53,366 --> 00:02:54,366
But there we go.

58
00:02:54,366 --> 00:02:57,266
What we need right now
is this cell and nothing else.

59
00:02:57,266 --> 00:03:00,266
And we actually don't have anything
to change inside.

60
00:03:00,366 --> 00:03:01,666
So that's all good.

61
00:03:01,666 --> 00:03:05,266
Now let's go back to our copy
of our natural

62
00:03:05,266 --> 00:03:07,133
language Processing implementation.

63
00:03:07,133 --> 00:03:09,266
Let's create a new code cell here.

64
00:03:09,266 --> 00:03:12,133
And let's just paste that cell to train.

65
00:03:12,133 --> 00:03:16,133
Well the Gaussian
naive base model on the training set

66
00:03:16,133 --> 00:03:20,066
composed of X train and Y train
that was just created just before.

67
00:03:20,566 --> 00:03:21,300
All right.

68
00:03:21,300 --> 00:03:21,800
Good.

69
00:03:21,800 --> 00:03:24,400
Now next step predicting the test results.

70
00:03:24,400 --> 00:03:27,166
Well, once again here
we won't have anything to do

71
00:03:27,166 --> 00:03:31,866
except a simple copy paste
still from our Naive Bayes implementation.

72
00:03:32,233 --> 00:03:33,200
Because you know what?

73
00:03:33,200 --> 00:03:38,600
We just want to do here is display next
to each other the vector of predictions

74
00:03:38,600 --> 00:03:42,400
and the vector of real results
containing the real reviews.

75
00:03:42,400 --> 00:03:45,400
You know, the real outcomes
of the reviews, whether they're positive,

76
00:03:45,433 --> 00:03:48,566
which gives a one or negative
which gives a zero.

77
00:03:48,900 --> 00:03:55,200
So here I just copied and I'm about
to paste that here in a new code cell.

78
00:03:55,600 --> 00:03:56,066
All right.

79
00:03:56,066 --> 00:03:59,100
And this will indeed
print next to each other.

80
00:03:59,100 --> 00:04:03,200
First the vector of predictions which
we got here in this first line of code,

81
00:04:03,466 --> 00:04:08,133
and the vector of real results containing
the real outcomes of the reviews.

82
00:04:08,566 --> 00:04:09,233
Perfect.

83
00:04:09,233 --> 00:04:11,700
And finally making the confusion matrix.

84
00:04:11,700 --> 00:04:13,166
That's our last step here.

85
00:04:13,166 --> 00:04:16,100
And once again
we're going to go back to our two kids.

86
00:04:16,100 --> 00:04:18,333
You know the classification toolkit
for Naive Bayes.

87
00:04:18,333 --> 00:04:20,666
We're going to scroll down a bit more.

88
00:04:20,666 --> 00:04:23,866
And we're going to find indeed
the confusion matrix

89
00:04:24,300 --> 00:04:27,533
computing as well, the accuracy score.

90
00:04:27,600 --> 00:04:31,266
You know, the accuracy being simply
the number of correct predictions

91
00:04:31,433 --> 00:04:34,966
divided by the total number
of observations in the test set.

92
00:04:34,966 --> 00:04:35,866
Of course.

93
00:04:35,866 --> 00:04:36,166
All right.

94
00:04:36,166 --> 00:04:40,000
So let's copy
and paste that in this new code cell.

95
00:04:40,000 --> 00:04:41,700
And there you go my friends.

96
00:04:41,700 --> 00:04:43,800
Now this implementation is over.

97
00:04:43,800 --> 00:04:46,066
We finished it in just a few seconds

98
00:04:46,066 --> 00:04:48,233
or you know in a few minutes
with the explanation.

99
00:04:48,233 --> 00:04:49,200
But there we go.

100
00:04:49,200 --> 00:04:52,466
That's what I mean
by juggling with your different toolkits.

101
00:04:52,500 --> 00:04:56,266
You can be super efficient
at implementing a classification

102
00:04:56,266 --> 00:04:59,533
or regression model
by using your code templates.

103
00:05:00,233 --> 00:05:00,933
All right.

104
00:05:00,933 --> 00:05:02,066
Now it's showtime.

105
00:05:02,066 --> 00:05:06,100
We will execute the cells
that we haven't executed so far.

106
00:05:06,100 --> 00:05:10,200
So the last one we executed with this one,
basically everything that is related

107
00:05:10,200 --> 00:05:11,800
to the bag of words model.

108
00:05:11,800 --> 00:05:16,100
So now let's play
the rest of the cell starting this one

109
00:05:16,100 --> 00:05:19,800
which will split the data
set into the training set and test set.

110
00:05:20,000 --> 00:05:20,866
All good.

111
00:05:20,866 --> 00:05:22,200
Now that we have the training set,

112
00:05:22,200 --> 00:05:25,400
we're going to train the Naive Bayes
model on the training set.

113
00:05:25,400 --> 00:05:26,900
And all good again.

114
00:05:26,900 --> 00:05:27,533
And now we're

115
00:05:27,533 --> 00:05:31,066
going to predict the test set results
by displaying next to each other

116
00:05:31,333 --> 00:05:36,566
the vector of predictions and the vector
of real outcomes of the reviews.

117
00:05:37,000 --> 00:05:38,866
And well, well look at this.

118
00:05:38,866 --> 00:05:43,200
We don't start well because we start
with three incorrect predictions.

119
00:05:43,366 --> 00:05:44,066
Right?

120
00:05:44,066 --> 00:05:47,833
For the first review, which, be
careful, is not well, love this place

121
00:05:47,833 --> 00:05:51,500
because these are the reviews
of the test set, not the whole data set.

122
00:05:51,500 --> 00:05:53,233
So this is not the first review.

123
00:05:53,233 --> 00:05:54,000
I love this place.

124
00:05:54,000 --> 00:05:55,033
This is just a random

125
00:05:55,033 --> 00:05:58,800
review taken from the original data set
and put into the test set.

126
00:05:59,100 --> 00:06:02,800
But anyway, for this first review
of the test set, our model predicted

127
00:06:02,800 --> 00:06:06,700
this review to be positive,
whereas in reality it is negative.

128
00:06:06,933 --> 00:06:10,766
Same for the second review predicted
positive, but in reality negative.

129
00:06:10,766 --> 00:06:14,466
The same for the third review predicted
positive, but in reality negative.

130
00:06:14,733 --> 00:06:16,400
And then this is correct.

131
00:06:16,400 --> 00:06:19,500
That's negative review
which was indeed predicted as negative.

132
00:06:19,633 --> 00:06:20,500
Same for this one.

133
00:06:20,500 --> 00:06:22,933
Negative review
predicted as negative here.

134
00:06:22,933 --> 00:06:24,066
Another mistake.

135
00:06:24,066 --> 00:06:26,966
Negative review
predicted as positive here.

136
00:06:26,966 --> 00:06:28,000
Correct prediction.

137
00:06:28,000 --> 00:06:31,500
Positive review predicted as positive
then incorrect prediction.

138
00:06:31,500 --> 00:06:35,433
Negative review predicted as positive,
then an incorrect prediction again.

139
00:06:35,433 --> 00:06:38,266
Negative review
predicted as positive and correct.

140
00:06:38,266 --> 00:06:40,166
Correct. Correct. Incorrect.

141
00:06:40,166 --> 00:06:41,766
Anyway. So yeah.

142
00:06:41,766 --> 00:06:45,200
But by scrolling down
we can see you know that

143
00:06:45,200 --> 00:06:48,666
we actually have many correct predictions.

144
00:06:49,000 --> 00:06:52,800
And anyway, we're going to check that
right away with our confusion

145
00:06:52,800 --> 00:06:53,966
matrix below.

146
00:06:53,966 --> 00:06:56,600
So I'm actually going to scroll down
from here.

147
00:06:56,600 --> 00:06:58,966
There we go. And perfect.

148
00:06:58,966 --> 00:07:02,800
So this is our last cell
which will display the confusion matrix

149
00:07:02,800 --> 00:07:06,300
and compute the accuracy score
which I will want you to beat

150
00:07:06,300 --> 00:07:07,633
right after this tutorial.

151
00:07:07,633 --> 00:07:09,833
As a final exercise of NLP.

152
00:07:09,833 --> 00:07:14,533
And so let's play the cell to see
what the confusion matrix looks like,

153
00:07:14,533 --> 00:07:19,666
and mostly to see the final accuracy,
which is 73% all right.

154
00:07:19,666 --> 00:07:20,833
So that's pretty good.

155
00:07:20,833 --> 00:07:22,800
But I'm sure we can do better.

156
00:07:22,800 --> 00:07:24,666
You know there are many ways to better.

157
00:07:24,666 --> 00:07:27,600
And so I really look forward
to seeing your results.

158
00:07:27,600 --> 00:07:31,100
You know after you experiment
with more classification models or even

159
00:07:31,333 --> 00:07:34,800
by doing a better cleaning of the text,
you know, the reviews,

160
00:07:34,966 --> 00:07:39,066
maybe you can add some more exclusions
in the list of stopwords.

161
00:07:39,066 --> 00:07:40,166
You know, we exclude it

162
00:07:40,166 --> 00:07:43,700
not from the list of stopwords,
but maybe you can exclude as well.

163
00:07:43,833 --> 00:07:47,600
Isn't you know, I know that isn't
is actually part of the Stopwords list.

164
00:07:47,766 --> 00:07:50,900
So you know you can do some extra work
in order to improve this,

165
00:07:51,100 --> 00:07:55,100
because I'm sure that we can get
a better accuracy than 73%.

166
00:07:55,100 --> 00:07:56,700
But still, this is pretty good.

167
00:07:56,700 --> 00:08:00,600
You know, remember that we actually
trained a machine to understand English.

168
00:08:00,800 --> 00:08:03,333
And, you know, a couple of years ago,
maybe decades ago,

169
00:08:03,333 --> 00:08:05,533
this would definitely seem
very challenging.

170
00:08:05,533 --> 00:08:07,633
But here we did it in just a few minutes.

171
00:08:07,633 --> 00:08:09,433
And so that's absolutely incredible.

172
00:08:09,433 --> 00:08:12,766
And the model predicts
whether these English written reviews

173
00:08:12,766 --> 00:08:16,933
are positive or negative
correctly 73% of the time.

174
00:08:16,933 --> 00:08:18,133
So that's really, really good.

175
00:08:18,133 --> 00:08:23,933
And that's the confusion matrix with 55
correct predictions of negative reviews,

176
00:08:23,933 --> 00:08:29,400
91 correct predictions of positive
reviews, 42 incorrect predictions

177
00:08:29,400 --> 00:08:33,800
of positive reviews, and 12 incorrect
predictions of negative reviews.

178
00:08:33,833 --> 00:08:37,266
All right,
so try to reduce these two numbers here

179
00:08:37,366 --> 00:08:39,900
and let me know
or let everybody know in the comments

180
00:08:39,900 --> 00:08:43,466
what you managed to get, which solution
you managed to improve.

181
00:08:43,466 --> 00:08:46,666
And I look forward to seeing
if you managed to, you know,

182
00:08:46,666 --> 00:08:48,566
for example, go over 80%.

183
00:08:48,566 --> 00:08:50,000
That would be fantastic.

184
00:08:50,000 --> 00:08:53,100
You know, you would definitely beat me
by pretty far.

185
00:08:53,700 --> 00:08:54,200
All right.

186
00:08:54,200 --> 00:08:57,066
So now we're done
with natural language processing.

187
00:08:57,066 --> 00:09:00,733
I hope you liked this introduction
to sentiment analysis.

188
00:09:00,966 --> 00:09:03,033
If you liked NLP and if you want to,

189
00:09:03,033 --> 00:09:06,666
you know, study more in-depth
this branch of machine learning,

190
00:09:06,766 --> 00:09:10,500
well know that we have other courses
about chatbots, about the Bert model.

191
00:09:10,633 --> 00:09:12,433
So I really recommend to check it out.

192
00:09:12,433 --> 00:09:17,266
But first I recommend that, you know you
complete this journey of machine learning.

193
00:09:17,266 --> 00:09:20,333
And speaking of this, well,
the next step of our journey here

194
00:09:20,333 --> 00:09:25,133
is to enter the fascinating world
of deep learning, where we will, you know,

195
00:09:25,133 --> 00:09:30,133
mimic the process of the human brain
to actually give for the first time

196
00:09:30,133 --> 00:09:35,533
to our AI, an artificial brain which will
itself perform some predictions.

197
00:09:35,700 --> 00:09:37,233
It's super fascinating.

198
00:09:37,233 --> 00:09:40,233
It's actually, you know, the most
fascinating branch of machine learning,

199
00:09:40,233 --> 00:09:43,800
because this is the closest one
to human intelligence.

200
00:09:44,066 --> 00:09:47,166
So now I just can't wait to see you
in this next

201
00:09:47,166 --> 00:09:50,166
part to enter the world of deep learning.

202
00:09:50,300 --> 00:09:52,366
And until then, enjoy machine learning.