﻿1
00:00:00,133 --> 00:00:02,400
Hello and welcome to this art tutorial.

2
00:00:02,400 --> 00:00:05,300
So in the previous section,
we implemented the apriori model

3
00:00:05,300 --> 00:00:09,366
to find some relevant association rules
that could help us find a more

4
00:00:09,366 --> 00:00:13,500
strategic placement of some products
in a grocery store in the South of France.

5
00:00:13,900 --> 00:00:17,466
And therefore, thanks to these association
rules, the manager of this store

6
00:00:17,466 --> 00:00:20,833
is likely to optimize the sales
and therefore increase the revenues.

7
00:00:21,266 --> 00:00:24,200
And accordingly,
we managed to create some added values

8
00:00:24,200 --> 00:00:27,566
in this business
thanks to association rule learning.

9
00:00:27,600 --> 00:00:31,600
And today we are going to implement
a new model of association

10
00:00:31,600 --> 00:00:34,666
rule learning,
which is called the E-Class model.

11
00:00:35,133 --> 00:00:39,500
So the first thing I want to say is to
prevent any disappointment you might have.

12
00:00:39,700 --> 00:00:43,900
The E-Class model is actually very simple
compared to what we did previously,

13
00:00:43,900 --> 00:00:47,300
because the model is basically
the apriori model,

14
00:00:47,333 --> 00:00:50,400
simplified
because previously we had two parameters.

15
00:00:50,400 --> 00:00:54,133
We had the support and the confidence,
and we also had the lift, by the way,

16
00:00:54,133 --> 00:00:57,466
when we ranked our rules
by their decreasing lifts.

17
00:00:57,900 --> 00:00:59,066
But here in the E-Class

18
00:00:59,066 --> 00:01:02,800
we will only have one parameter
which will be the support parameter.

19
00:01:03,033 --> 00:01:06,766
And therefore when we obtain our rules,
that will not be the rules

20
00:01:06,766 --> 00:01:07,833
that we obtained earlier.

21
00:01:07,833 --> 00:01:11,966
Like people who bought this also,
but that but we will just get some

22
00:01:11,966 --> 00:01:15,300
different sets of products frequently,
but together we will get

23
00:01:15,466 --> 00:01:16,833
the different sets of products.

24
00:01:16,833 --> 00:01:20,700
There are the most frequently purchased
together, so that's what we'll get.

25
00:01:20,733 --> 00:01:22,366
We have to expect this.

26
00:01:22,366 --> 00:01:26,300
But you know, the exam model can be
very useful if you don't have much time.

27
00:01:26,333 --> 00:01:29,066
If you want to get some simple results,

28
00:01:29,066 --> 00:01:29,800
if you don't want to play

29
00:01:29,800 --> 00:01:33,066
with too many parameters like the support
and the confidence here,

30
00:01:33,100 --> 00:01:35,500
we won't even have to compute a value
for the support.

31
00:01:35,500 --> 00:01:37,433
And moreover, we won't even have to

32
00:01:37,433 --> 00:01:40,833
choose a value for the confidence
because there is no confidence parameter.

33
00:01:41,166 --> 00:01:41,933
So you'll see

34
00:01:41,933 --> 00:01:45,833
this is the way of using association
rule learning in the simplest way.

35
00:01:46,000 --> 00:01:47,333
So let's do it right now.

36
00:01:47,333 --> 00:01:49,933
You'll see what I mean
at the end of this tutorial.

37
00:01:49,933 --> 00:01:52,166
So let's start with the very first step.

38
00:01:52,166 --> 00:01:54,233
Setting
the right folder is working directory.

39
00:01:54,233 --> 00:01:56,433
So we'll go to our Machine
Learning A-Z folder.

40
00:01:56,433 --> 00:01:58,800
Part five Association Rule Learning.

41
00:01:58,800 --> 00:02:01,033
And right now we are in a club.

42
00:02:01,033 --> 00:02:02,233
So here it is.

43
00:02:02,233 --> 00:02:04,200
We have the market basket optimization.

44
00:02:04,200 --> 00:02:06,266
We will work on the same business problem.

45
00:02:06,266 --> 00:02:08,633
And so we can click
on this more button here.

46
00:02:08,633 --> 00:02:10,866
And now set as working directory.

47
00:02:10,866 --> 00:02:15,366
And you know since I told you
the model is a simplified version

48
00:02:15,366 --> 00:02:19,733
of the apriori model, what we'll do now
is take our apriori model

49
00:02:20,100 --> 00:02:23,100
and you'll see that
it will almost be the same.

50
00:02:23,133 --> 00:02:25,300
And we will just need to change one thing.

51
00:02:25,300 --> 00:02:28,300
So I'm going to select everything
from here

52
00:02:29,100 --> 00:02:31,600
to here copy.

53
00:02:31,600 --> 00:02:34,200
And I'm going to paste it in the model.

54
00:02:34,200 --> 00:02:35,300
All right.

55
00:02:35,300 --> 00:02:38,100
So here in this section
data preprocessing.

56
00:02:38,100 --> 00:02:39,800
We don't have anything to change.

57
00:02:39,800 --> 00:02:45,333
We are just going to import the data
set with the read transaction functions.

58
00:02:45,333 --> 00:02:47,700
Be careful not with the red dot CSV file.

59
00:02:47,700 --> 00:02:50,700
Because we still need to have
our sparse matrix.

60
00:02:50,933 --> 00:02:54,000
That will also be
an input of the function,

61
00:02:54,233 --> 00:02:56,400
like here for the apriori function.

62
00:02:56,400 --> 00:02:59,800
So we will select this and execute.

63
00:03:00,266 --> 00:03:03,266
And of course we get the same number
of duplicates.

64
00:03:03,333 --> 00:03:05,500
The CSV file hasn't changed.

65
00:03:05,500 --> 00:03:08,600
We have five duplicates and no duplicates.

66
00:03:08,600 --> 00:03:10,700
Just five duplicates okay.

67
00:03:10,700 --> 00:03:11,200
Then of course

68
00:03:11,200 --> 00:03:14,600
we can use the summary function
to get some info about this data set.

69
00:03:14,600 --> 00:03:17,600
But of course these are going to be
the same enforce before.

70
00:03:17,633 --> 00:03:20,766
So we have 7500 transactions

71
00:03:21,100 --> 00:03:26,300
119 products, and the density is 0.03.

72
00:03:26,300 --> 00:03:31,900
That means that the proportion of non-zero
values in the matrix is 0.03 3%.

73
00:03:32,333 --> 00:03:34,433
And then of course we have
the most frequent items.

74
00:03:34,433 --> 00:03:36,666
Mineral water comes first and eggs.

75
00:03:36,666 --> 00:03:39,600
And we can actually have a better
look at these

76
00:03:39,600 --> 00:03:43,066
most frequent items
by selecting this line and execute.

77
00:03:43,400 --> 00:03:46,400
And of course
we get the same frequency plot.

78
00:03:46,466 --> 00:03:48,133
So that's exactly like before.

79
00:03:48,133 --> 00:03:49,866
We don't have anything to change here.

80
00:03:49,866 --> 00:03:53,666
However, now we enter
the second code section which is to train

81
00:03:53,933 --> 00:03:55,666
the model on the data set.

82
00:03:55,666 --> 00:04:00,300
So first we will just replace
apriori here by Eclat.

83
00:04:00,966 --> 00:04:02,066
Here we go.

84
00:04:02,066 --> 00:04:04,566
And guess what? It is so simple.

85
00:04:04,566 --> 00:04:08,166
Here we have the apriori function
to train the apriori model.

86
00:04:08,466 --> 00:04:11,466
And guess what the function is going to be
to train the model.

87
00:04:11,600 --> 00:04:14,966
It's going to be a class. So very simple.

88
00:04:15,000 --> 00:04:18,000
We are almost ready to train it
on our data set.

89
00:04:18,066 --> 00:04:21,433
But of course
the economical is much more simple.

90
00:04:21,633 --> 00:04:24,366
It doesn't include the confidence
in the parameters.

91
00:04:24,366 --> 00:04:28,066
If right now we get to the slide
of the Eclat algorithm from the intuition

92
00:04:28,066 --> 00:04:32,466
tutorial, well, we can see that
the algorithm has three steps.

93
00:04:32,766 --> 00:04:35,366
And the first step
is to set a minimum support.

94
00:04:35,366 --> 00:04:38,433
Remember before, for the apriori
algorithm, the first step

95
00:04:38,433 --> 00:04:41,433
was to set a minimum support
and a minimum confidence.

96
00:04:41,500 --> 00:04:45,133
Now we only need to set a minimum support
so we don't need this.

97
00:04:45,333 --> 00:04:47,966
Actually this parameter
is only for the apriori model.

98
00:04:47,966 --> 00:04:49,900
So if we keep that we will get an error.

99
00:04:51,033 --> 00:04:52,933
So I'll just remove this and

100
00:04:52,933 --> 00:04:56,666
we can leave the minimum support to 0.04.

101
00:04:56,666 --> 00:04:59,466
But you will see that
it's even not necessary here.

102
00:04:59,466 --> 00:05:00,933
You'll see why at the end.

103
00:05:00,933 --> 00:05:03,933
However
we might need to add another parameter.

104
00:05:04,300 --> 00:05:08,066
It's because since, you know,
I mentioned that the algorithm

105
00:05:08,066 --> 00:05:12,900
is just going to return the sets of items
most frequently bought together.

106
00:05:13,200 --> 00:05:17,000
Well, it wouldn't be interesting
to have sets of only one item.

107
00:05:17,300 --> 00:05:18,733
So in order to get some sets

108
00:05:18,733 --> 00:05:22,500
of at least two items,
we will add another parameter here,

109
00:05:22,566 --> 00:05:27,466
which we actually encountered earlier
and which is the Midlane parameter.

110
00:05:27,733 --> 00:05:31,900
And we will set it to two,
because we want to have the different sets

111
00:05:31,900 --> 00:05:35,433
of at least two items
most frequently purchased together.

112
00:05:35,933 --> 00:05:37,866
Okay. And now we're ready.

113
00:05:37,866 --> 00:05:39,366
See how it was? Simple.

114
00:05:39,366 --> 00:05:44,300
So let's select this
and train the model on our data set.

115
00:05:45,333 --> 00:05:46,266
Done.

116
00:05:46,266 --> 00:05:47,266
Okay.

117
00:05:47,266 --> 00:05:49,700
So now things change a little bit.

118
00:05:49,700 --> 00:05:53,900
First of course we see that
we just trained the model obviously.

119
00:05:54,300 --> 00:05:56,700
Then we have the parameter specification
like before.

120
00:05:56,700 --> 00:05:59,100
But this time
we don't have the confidence parameter.

121
00:05:59,100 --> 00:06:02,966
We have the support that we set to 0.04.

122
00:06:02,966 --> 00:06:04,600
That is the minimum support.

123
00:06:04,600 --> 00:06:07,533
And of course this time
we have the middle and equals to two.

124
00:06:07,533 --> 00:06:10,500
Remember before
the million parameter was set to one.

125
00:06:10,500 --> 00:06:13,933
But we didn't have any problem with that
because all our rules contained

126
00:06:13,933 --> 00:06:15,633
at least two products.

127
00:06:15,633 --> 00:06:19,333
But when using class we need to set this
Midlane parameter to two,

128
00:06:19,566 --> 00:06:23,433
because otherwise
we will get some sets of only one item.

129
00:06:24,066 --> 00:06:24,600
Okay.

130
00:06:24,600 --> 00:06:27,600
Then we have these other more
advanced informations here.

131
00:06:27,800 --> 00:06:30,866
And then there is something
that I would like to highlight here.

132
00:06:31,200 --> 00:06:34,800
It's the number of sets and not rules.

133
00:06:34,800 --> 00:06:38,833
Remember before we had
you know, let's say we had 845,

134
00:06:38,900 --> 00:06:43,766
it was written 845 rules
because we had the rules of the forum.

135
00:06:44,100 --> 00:06:45,566
If people buy live cream,

136
00:06:45,566 --> 00:06:48,900
then they're likely to buy chicken
with a confidence of 40%.

137
00:06:48,900 --> 00:06:51,033
That is, with 40% chance.

138
00:06:51,033 --> 00:06:54,900
And here, since, you know, we don't get
these kind of rules of this form

139
00:06:55,166 --> 00:07:00,600
and we only get two sets of items, well,
this time we have indeed 845 sets.

140
00:07:00,933 --> 00:07:04,033
So even if Eclair is considered
as a association

141
00:07:04,033 --> 00:07:07,100
rule learning model,
well this doesn't return rules.

142
00:07:07,100 --> 00:07:09,533
This actually returns some sets.

143
00:07:09,533 --> 00:07:11,700
Okay.
So we're going to have a look at the sets.

144
00:07:11,700 --> 00:07:12,433
Right now.

145
00:07:12,433 --> 00:07:16,200
We are ready to move
to the last step of our model.

146
00:07:16,200 --> 00:07:18,300
Well we can actually jump
back to the slide.

147
00:07:18,300 --> 00:07:22,166
And as we can see the step two is to take
all the subsets and transactions

148
00:07:22,166 --> 00:07:24,300
having higher support
than minimum support.

149
00:07:24,300 --> 00:07:26,800
So that's what the function
excluded itself.

150
00:07:26,800 --> 00:07:31,666
And then the last step step three is
to sort this subset by decreasing support.

151
00:07:31,800 --> 00:07:35,266
So this time it's not by decreasing lift
like for Priory

152
00:07:35,566 --> 00:07:37,000
there is no lift in Aguila.

153
00:07:37,000 --> 00:07:41,033
So this time we're going to
sort the rules by decreasing

154
00:07:42,300 --> 00:07:44,100
support.

155
00:07:44,100 --> 00:07:44,566
All right.

156
00:07:44,566 --> 00:07:47,966
And we're going to take the ten
first rules with the ten highest support.

157
00:07:48,300 --> 00:07:49,733
So we are actually ready.

158
00:07:49,733 --> 00:07:52,433
Well the whole code is actually ready.

159
00:07:52,433 --> 00:07:53,666
We did it very efficiently.

160
00:07:53,666 --> 00:07:55,366
Well it was very simple.

161
00:07:55,366 --> 00:07:57,900
So eventually
let's have a look at the rules.

162
00:07:57,900 --> 00:08:01,500
And you'll see that it's much more simple
and honestly less

163
00:08:01,500 --> 00:08:03,600
interesting than apriori.

164
00:08:03,600 --> 00:08:05,033
But that's what we'll get.

165
00:08:05,033 --> 00:08:08,600
So I'm going to select this line
and execute.

166
00:08:09,033 --> 00:08:13,466
And as I just told you,
we simply get the different sets of items

167
00:08:13,833 --> 00:08:17,200
most frequently purchased together.

168
00:08:17,500 --> 00:08:21,300
So for example, the set of items
most frequently purchased together

169
00:08:21,300 --> 00:08:26,633
is mineral water and spaghetti
with the support of 0.059.

170
00:08:27,066 --> 00:08:30,366
And then it's chocolate,
and then it's chocolate and mineral water,

171
00:08:30,566 --> 00:08:34,733
then eggs and mineral water and that's,
of course, strongly related to this

172
00:08:35,133 --> 00:08:38,133
most frequently
purchased products in this store.

173
00:08:38,133 --> 00:08:41,566
So that's the éclair results,
much less interesting than the apriori.

174
00:08:41,866 --> 00:08:45,800
But that can be very useful if you're
looking for some very simple information.

175
00:08:46,066 --> 00:08:50,533
And by the way, since I was telling you,
if we, you know, change the support

176
00:08:50,533 --> 00:08:55,266
and set it to 0.03 like we did
for the apriori model in the first place,

177
00:08:55,500 --> 00:08:56,400
well, you'll see that

178
00:08:56,400 --> 00:08:59,966
we'll get the same ranking here
because as you can see, the supports

179
00:09:00,133 --> 00:09:03,966
of these ten first sets of items
most frequently purchase together,

180
00:09:04,200 --> 00:09:08,733
all have supports
higher than 0.04 or 0.03.

181
00:09:09,200 --> 00:09:13,766
So indeed, if we retrain the model
with the minimum support of 0.03,

182
00:09:14,533 --> 00:09:17,533
and if we select that again and execute,

183
00:09:17,733 --> 00:09:22,500
we get the same ranking with mineral water
and spaghetti coming first

184
00:09:22,833 --> 00:09:25,833
and then frozen vegetables, mineral water

185
00:09:26,100 --> 00:09:29,700
as the 10th set of products
most frequently purchased together.

186
00:09:30,333 --> 00:09:30,733
All right.

187
00:09:30,733 --> 00:09:33,200
So that was the Éclair model.

188
00:09:33,200 --> 00:09:35,300
Again,
if you want to have a serious analysis

189
00:09:35,300 --> 00:09:38,333
about how to create some added value
for your business,

190
00:09:38,600 --> 00:09:42,266
optimize the sales and the revenue,
you should go for apriori.

191
00:09:42,600 --> 00:09:45,733
But if you are looking
for some very simple informations like

192
00:09:45,733 --> 00:09:48,733
the sets of products
most frequently purchased together,

193
00:09:48,833 --> 00:09:51,266
then you can go for it.

194
00:09:51,266 --> 00:09:52,466
So congratulations!

195
00:09:52,466 --> 00:09:56,833
Anyway, you now know how to implement
two association rule learning models

196
00:09:57,000 --> 00:09:58,633
the apriori and Eclat.

197
00:09:58,633 --> 00:10:01,033
And you know what to use them for.

198
00:10:01,033 --> 00:10:03,500
So thank you very much
for watching these tutorials

199
00:10:03,500 --> 00:10:05,700
and I look forward to seeing you
in the next part.

200
00:10:05,700 --> 00:10:07,433
Reinforcement learning.

201
00:10:07,433 --> 00:10:09,166
Until then, enjoy machine learning.
