1
00:00:00,100 --> 00:00:01,100
Hello my friends!

2
00:00:01,100 --> 00:00:03,466
I'm super excited to start now.

3
00:00:03,466 --> 00:00:07,566
A part of this course
that we've been excitedly waiting for.

4
00:00:07,766 --> 00:00:10,300
I'm talking,
of course, about deep learning.

5
00:00:10,300 --> 00:00:13,200
You just got the amazing intuition
lectures by Kirill.

6
00:00:13,200 --> 00:00:16,366
And so now you understand that
what deep learning consists

7
00:00:16,366 --> 00:00:19,366
of is building an artificial brain.

8
00:00:19,400 --> 00:00:20,666
So there you go, my friends.

9
00:00:20,666 --> 00:00:26,633
Now we're about to build an artificial
brain using the brand new TensorFlow 2.0.

10
00:00:26,966 --> 00:00:28,400
This will be super exciting.

11
00:00:28,400 --> 00:00:32,266
We will really build a deep neural network
with neurons

12
00:00:32,266 --> 00:00:34,933
and fully connected layers
connecting these neurons.

13
00:00:34,933 --> 00:00:38,333
And we will apply all this
to a business problem as usual.

14
00:00:38,333 --> 00:00:40,233
And now that we are,
you know, quite advanced.

15
00:00:40,233 --> 00:00:41,000
And of course

16
00:00:41,000 --> 00:00:44,100
the data set we'll be working on
looks more like a real world

17
00:00:44,100 --> 00:00:47,400
data set with many observations
and many features.

18
00:00:47,400 --> 00:00:48,066
And you will see that

19
00:00:48,066 --> 00:00:51,766
we will actually have to not only use
our data preprocessing template,

20
00:00:51,933 --> 00:00:55,500
but also use some of the tools of our data
preprocessing toolkit.

21
00:00:56,000 --> 00:00:57,600
So are you ready?

22
00:00:57,600 --> 00:00:59,333
Are you ready to go next level?

23
00:00:59,333 --> 00:00:59,833
Indeed.

24
00:00:59,833 --> 00:01:02,833
Now we're about to do some more advanced
machine learning.

25
00:01:02,900 --> 00:01:04,233
And so I'm super excited

26
00:01:04,233 --> 00:01:07,833
that you're even pushing further
your expertise in machine learning.

27
00:01:08,300 --> 00:01:09,300
All right, so let's do this.

28
00:01:09,300 --> 00:01:12,700
But first let's just make sure
everyone here is on the same page.

29
00:01:12,700 --> 00:01:15,400
So this is the folder
of the whole machinery.

30
00:01:15,400 --> 00:01:17,666
It is it with all the codes and data sets.

31
00:01:17,666 --> 00:01:20,266
I gave you the link to this folder
right before this tutorial.

32
00:01:20,266 --> 00:01:21,900
So make sure to connect to it.

33
00:01:21,900 --> 00:01:23,066
And now there we go.

34
00:01:23,066 --> 00:01:26,433
Let's end to part eight Deep learning with

35
00:01:26,433 --> 00:01:29,933
first
the classic artificial neural network,

36
00:01:29,933 --> 00:01:34,700
meaning the fully connected neural network
with only fully connected layers.

37
00:01:34,700 --> 00:01:38,566
You know, with no convolutional layers
or other types of layers.

38
00:01:38,600 --> 00:01:42,166
Here we will just have an input vector
containing different features,

39
00:01:42,166 --> 00:01:45,833
and we will predict an outcome
which will be a binary variable,

40
00:01:45,866 --> 00:01:48,066
because you have to know
that actually artificial neural

41
00:01:48,066 --> 00:01:51,733
networks can be both used for regression
or classification.

42
00:01:51,966 --> 00:01:54,800
And here we're going to do it
for classification.

43
00:01:54,800 --> 00:01:58,500
However note that we have a free course
on artificial neural network,

44
00:01:58,666 --> 00:02:02,600
in which this time we built an artificial
neural network for regression.

45
00:02:02,600 --> 00:02:03,800
So make sure to check it out.

46
00:02:03,800 --> 00:02:06,000
I will think of
including the links somewhere

47
00:02:06,000 --> 00:02:08,100
so that you can get it as well,
but it's really good.

48
00:02:08,100 --> 00:02:09,600
You will have both cases.

49
00:02:09,600 --> 00:02:13,433
You know, the classification case
with this course and the regression case

50
00:02:13,433 --> 00:02:15,500
with the other free course. All right.

51
00:02:15,500 --> 00:02:18,600
So now as usual
we're going to start with Python.

52
00:02:18,600 --> 00:02:19,733
And there we go.

53
00:02:19,733 --> 00:02:24,666
This folder contains the implementation
first artificial neural network that ipynb

54
00:02:24,900 --> 00:02:28,666
which you can either open with Jupyter
notebook or Google Collaboratory.

55
00:02:29,000 --> 00:02:33,500
And we have of course our data set,
which I'm going to explain right now.

56
00:02:34,133 --> 00:02:34,500
All right.

57
00:02:34,500 --> 00:02:35,500
So as you can see,

58
00:02:35,500 --> 00:02:40,133
as I told you, it looks indeed
more like a real world data set, right?

59
00:02:40,133 --> 00:02:40,866
For the first time.

60
00:02:40,866 --> 00:02:44,200
Our data set takes the full screen here
because indeed, this time

61
00:02:44,200 --> 00:02:48,433
we have many features, you know,
starting from here up to this one.

62
00:02:48,433 --> 00:02:51,033
And the dependent variable right here.

63
00:02:51,033 --> 00:02:51,400
All right.

64
00:02:51,400 --> 00:02:53,600
So let me explain what this is about.

65
00:02:53,600 --> 00:02:56,600
This is the data set of a bank

66
00:02:56,666 --> 00:03:00,533
which collected some informations
about their customers.

67
00:03:00,866 --> 00:03:03,166
These informations are well row number.

68
00:03:03,166 --> 00:03:05,100
That's just a non relevant feature.

69
00:03:05,100 --> 00:03:06,266
We will get rid of it.

70
00:03:06,266 --> 00:03:07,466
Then the customer ID.

71
00:03:07,466 --> 00:03:10,566
That's just the identification
key of each customer.

72
00:03:10,866 --> 00:03:15,000
The surname, the credit score,
the geography meaning the country,

73
00:03:15,000 --> 00:03:18,233
the customer lives
in, the gender, the age,

74
00:03:18,233 --> 00:03:21,600
the tenure meaning the number of years
that have been in the bank,

75
00:03:21,900 --> 00:03:25,433
the balance meaning the amount of money
they have on their account.

76
00:03:25,733 --> 00:03:27,833
The number of products
they use from the bank.

77
00:03:27,833 --> 00:03:29,400
You know, like a credit card

78
00:03:29,400 --> 00:03:34,133
or a checkbook or a Mastercard
or even a loan or home loan.

79
00:03:34,133 --> 00:03:36,000
You know, any banking products.

80
00:03:36,000 --> 00:03:36,400
Okay.

81
00:03:36,400 --> 00:03:39,266
So that's the number of products
each customer has.

82
00:03:39,266 --> 00:03:43,366
Then has credit card, yes or no, meaning
that this variable is equal to one.

83
00:03:43,366 --> 00:03:48,600
If the customer has a credit card and zero
otherwise is active member,

84
00:03:48,600 --> 00:03:52,866
meaning is the customer active,
is he or she using the bank,

85
00:03:52,866 --> 00:03:57,266
you know, connecting to its account or
using its credit card or any other card?

86
00:03:57,266 --> 00:03:58,133
You know, let's say

87
00:03:58,133 --> 00:04:01,866
they have some measurement system
to measure if a customer is active or not.

88
00:04:01,866 --> 00:04:02,566
And one means,

89
00:04:02,566 --> 00:04:06,600
of course, that the customer is active
and zero means that it is an active.

90
00:04:06,666 --> 00:04:09,300
Okay then estimated salary.

91
00:04:09,300 --> 00:04:13,400
Well, you know, the salary of the customer
estimated by the bank and that's it.

92
00:04:13,400 --> 00:04:14,800
That's the last feature.

93
00:04:14,800 --> 00:04:18,066
And then the last column here it is

94
00:04:18,066 --> 00:04:21,566
the dependent variable
and it tells if yes or no.

95
00:04:21,866 --> 00:04:26,100
Well the customer stayed in the bank
or left the bank.

96
00:04:26,366 --> 00:04:30,833
One means that it left the bank
as in exited equals yes.

97
00:04:31,000 --> 00:04:34,666
And zero means that the customer
stayed in the bank as an exit.

98
00:04:34,666 --> 00:04:35,766
It equals no.

99
00:04:35,766 --> 00:04:40,433
So what happened in reality
is that this bank actually observed

100
00:04:40,466 --> 00:04:43,966
their customers for a certain period
of time, let's say six months.

101
00:04:44,266 --> 00:04:48,600
They observed if during the six months
they left the bank or stayed in the bank,

102
00:04:48,733 --> 00:04:52,533
and they gathered these outcomes
in this last dependent variable,

103
00:04:52,800 --> 00:04:55,733
and at the same time, you know,
they got all these features.

104
00:04:55,733 --> 00:05:01,200
Well, to guess what, understand
the correlations between these features

105
00:05:01,400 --> 00:05:06,266
and the fact whether or not the customer
stays in the bank or leave the bank.

106
00:05:06,500 --> 00:05:07,966
And that makes sense, right?

107
00:05:07,966 --> 00:05:11,400
Because a bank wants to have
the maximum customers, right?

108
00:05:11,400 --> 00:05:12,300
That's how they make money.

109
00:05:12,300 --> 00:05:15,800
The more customers they have, the more
they have money in the bank, and the more

110
00:05:15,800 --> 00:05:19,566
they make money from the diverse products
they offer to their customers.

111
00:05:19,733 --> 00:05:23,533
So of course, their interest
is to keep the maximum customers,

112
00:05:23,733 --> 00:05:27,366
and therefore they made this data
set to understand

113
00:05:27,400 --> 00:05:31,033
the reasons
somehow why customers leave the bank.

114
00:05:31,300 --> 00:05:34,800
And mostly once
they managed to build a predictive model

115
00:05:34,800 --> 00:05:38,100
that can predict
if any new customer leaves the bank.

116
00:05:38,100 --> 00:05:41,100
You know, a model that was trained,
of course, on this data set.

117
00:05:41,133 --> 00:05:44,833
Well, they will deploy this model
on new customers

118
00:05:44,833 --> 00:05:47,933
and for all the customers,
where the model predicts that

119
00:05:47,933 --> 00:05:51,866
the customer leaves the bank, well,
they will be prepared and they might do

120
00:05:51,866 --> 00:05:56,666
some special offer to the customer
so that it will stay in the bank, you see.

121
00:05:57,000 --> 00:06:02,633
So all this is to prevent the maximum
customers from leaving the bank.

122
00:06:02,633 --> 00:06:04,733
And why is this girl churn modeling?

123
00:06:04,733 --> 00:06:07,900
Because customer churn
means exactly the situation

124
00:06:07,900 --> 00:06:11,833
where some customers exit,
you know, become no longer customer.

125
00:06:12,300 --> 00:06:12,900
All right.

126
00:06:12,900 --> 00:06:17,200
And of course, the bank has asked you
the most down to data scientist

127
00:06:17,200 --> 00:06:21,166
to make this predictive model,
to first train it on this data

128
00:06:21,166 --> 00:06:24,300
set to understand the correlations
of all these features

129
00:06:24,300 --> 00:06:29,300
and the dependent variable, and then
to deploy this model on future customers.

130
00:06:29,533 --> 00:06:32,600
And you will see that in the
implementation, we will actually deploy

131
00:06:32,866 --> 00:06:36,533
the future machine learning
model will get on a different customer,

132
00:06:36,533 --> 00:06:39,300
you know, not part of this data set,
so as to predict

133
00:06:39,300 --> 00:06:43,333
if this new customer will stay in the bank
or leaves the bank.

134
00:06:43,333 --> 00:06:44,700
And even better than this,

135
00:06:44,700 --> 00:06:49,500
we will actually predict the probability
that this customer leaves the bank.

136
00:06:49,833 --> 00:06:51,666
So we have a lot to do ahead of us.

137
00:06:51,666 --> 00:06:53,033
But I'm super excited

138
00:06:53,033 --> 00:06:56,333
because deep learning is a fascinating
branch of machine learning.

139
00:06:56,466 --> 00:07:00,966
So let's start right away
and let's open our implementation

140
00:07:00,966 --> 00:07:05,366
artificial Neural Network dot Ipynb,
which you can feel free to open

141
00:07:05,366 --> 00:07:07,200
with either Google Collaboratory

142
00:07:07,200 --> 00:07:10,966
as I'm about to do, or Jupyter Notebook
as you want.

143
00:07:11,133 --> 00:07:15,333
Just make sure you feel comfortable
coding on your favorite IDE.

144
00:07:16,366 --> 00:07:16,800
All right,

145
00:07:16,800 --> 00:07:20,700
so now the implementation is opening
laying it out the notebook.

146
00:07:20,700 --> 00:07:21,633
Perfect.

147
00:07:21,633 --> 00:07:24,600
That's the implementation once again
in read only mode.

148
00:07:24,600 --> 00:07:29,633
So right now we will click file here
to create a copy of this implementation.

149
00:07:29,900 --> 00:07:33,900
And as usual
we will re-implement all this from scratch

150
00:07:34,166 --> 00:07:37,366
so that we can really learn by doing.

151
00:07:37,900 --> 00:07:38,433
All right.

152
00:07:38,433 --> 00:07:39,300
So that's the copy.

153
00:07:39,300 --> 00:07:42,833
Now let's remove all the cells right.

154
00:07:42,966 --> 00:07:47,333
This one not the text cells of course,
because we want to keep the well

155
00:07:47,333 --> 00:07:49,833
highlighted structure
of this implementation.

156
00:07:49,833 --> 00:07:52,533
But let's definitely remove
the code cells.

157
00:07:52,533 --> 00:07:53,533
So there are many of them.

158
00:07:53,533 --> 00:07:56,133
So you will actually have
a bit of time doing it.

159
00:07:56,133 --> 00:07:58,666
But you know it's
not that long. It's fine.

160
00:07:58,666 --> 00:07:59,766
There is a long data

161
00:07:59,766 --> 00:08:04,100
preprocessing phase and then a long phase
of building the neural network.

162
00:08:04,366 --> 00:08:07,733
And I really gave the details
of the implementation.

163
00:08:07,733 --> 00:08:10,233
So you will actually see many steps.

164
00:08:10,233 --> 00:08:11,066
All right.

165
00:08:11,066 --> 00:08:14,700
And there are so many steps
that I actually added a level of

166
00:08:14,700 --> 00:08:18,666
structure because as you can see
the full implementation is divided

167
00:08:18,666 --> 00:08:20,433
into three parts. Right.

168
00:08:20,433 --> 00:08:22,500
There is a homework at the end.
The solution.

169
00:08:22,500 --> 00:08:24,966
So let me not show it to you.

170
00:08:24,966 --> 00:08:29,433
And that's almost the end
confusion matrix and perfect.

171
00:08:29,600 --> 00:08:33,900
All right okay so as you saw
this is a long implementation.

172
00:08:33,900 --> 00:08:35,400
This will be a long part.

173
00:08:35,400 --> 00:08:37,766
But it is absolutely worth it.

174
00:08:37,766 --> 00:08:42,500
Deep learning is one of the most powerful
branches of machine learning okay.

175
00:08:42,500 --> 00:08:43,466
So let's see.

176
00:08:43,466 --> 00:08:46,466
We start first
by importing the libraries as usual.

177
00:08:46,566 --> 00:08:49,433
And then we enter part
one data preprocessing.

178
00:08:49,433 --> 00:08:53,633
So that's the first part of the whole
implementation structured in four parts

179
00:08:53,633 --> 00:08:56,733
which are the following parts
one data preprocessing part

180
00:08:56,733 --> 00:08:59,833
two building two and part three training,

181
00:09:00,133 --> 00:09:03,933
and in part for making the predictions
and evaluating the model.

182
00:09:04,300 --> 00:09:07,000
And then each part
well we have two different steps.

183
00:09:07,000 --> 00:09:09,966
In part one data preprocessing
we will first import the data set.

184
00:09:09,966 --> 00:09:12,900
Of course then we'll have some data
preprocessing to do.

185
00:09:12,900 --> 00:09:17,233
You know, not only the classic steps
of the template but also some extra tools.

186
00:09:17,233 --> 00:09:18,866
And we will see that together.

187
00:09:18,866 --> 00:09:23,800
Then in part two we will first initialize
and then we'll add an input layer.

188
00:09:23,800 --> 00:09:26,533
And the first hidden layer
to our artificial brain.

189
00:09:26,533 --> 00:09:29,700
Then we'll add the second hidden layer,
then the output layer.

190
00:09:30,133 --> 00:09:33,833
Then in the next part three training
and we will first start

191
00:09:33,833 --> 00:09:37,833
by compiling the CNN to, you know,
an optimizer and a loss function.

192
00:09:38,166 --> 00:09:40,900
Then we will train the CNN
on the training set.

193
00:09:40,900 --> 00:09:45,166
And finally in part four, well,
we will deploy our model in production

194
00:09:45,166 --> 00:09:48,166
to predict the result
of a single observation, meaning

195
00:09:48,233 --> 00:09:51,900
to predict if a new customer will stay
or leave the bank,

196
00:09:52,333 --> 00:09:56,600
then we will predict the test results
to, you know, get that Y vector

197
00:09:56,700 --> 00:10:01,800
to eventually make the confusion matrix
with the final accuracy.

198
00:10:02,600 --> 00:10:03,133
All right.

199
00:10:03,133 --> 00:10:06,333
So once again as you can see
this implementation is pretty long.

200
00:10:06,333 --> 00:10:09,766
So make sure to have full energy
and full motivation for this

201
00:10:10,000 --> 00:10:13,733
because we have a long
but yet exciting journey in front of us.

202
00:10:13,900 --> 00:10:15,200
So as soon as you're ready,

203
00:10:15,200 --> 00:10:18,900
well, meet me in the next tutorial
to smash this implementation.

204
00:10:19,066 --> 00:10:20,900
And until then, enjoy machine learning.