1
00:00:00,200 --> 00:00:02,533
Hello and welcome to this art tutorial.

2
00:00:02,533 --> 00:00:05,533
So in the previous tutorials
we did the data pre-processing step.

3
00:00:05,533 --> 00:00:08,533
And then we train our apriori model
on the data set

4
00:00:08,800 --> 00:00:12,900
with a minimum support of 0.030.

5
00:00:12,900 --> 00:00:16,000
3% and a minimum confidence of 40%.

6
00:00:16,233 --> 00:00:19,933
And now we did the job, and we are finally
getting to the exciting step,

7
00:00:19,933 --> 00:00:24,333
which is to visualize the results,
that is, to look at the rules explicitly.

8
00:00:24,633 --> 00:00:27,633
We're going to have the list
of the strongest rules,

9
00:00:27,800 --> 00:00:32,133
and we will eventually know how to place
the products to optimize the sales.

10
00:00:32,533 --> 00:00:34,500
Okay, so let's do it without waiting.

11
00:00:34,500 --> 00:00:36,433
This will actually be very easy as well.

12
00:00:36,433 --> 00:00:38,100
We will only need one line.

13
00:00:38,100 --> 00:00:42,166
But before we write that line, let's jump
back to the apriori algorithm.

14
00:00:42,166 --> 00:00:42,866
Intuition.

15
00:00:42,866 --> 00:00:44,700
Slide two. See the steps

16
00:00:44,700 --> 00:00:48,000
we already did in the algorithm
and what steps we need to do left?

17
00:00:48,466 --> 00:00:51,666
Okay, so step one was to set the minimum
support in confidence.

18
00:00:51,900 --> 00:00:53,200
That was done here.

19
00:00:53,200 --> 00:00:58,200
We set the minimum support here of 0.3
percent and the minimum confidence of 40%.

20
00:00:58,500 --> 00:00:59,933
That's not our final word.

21
00:00:59,933 --> 00:01:02,933
We might need
to change the confidence again.

22
00:01:02,933 --> 00:01:05,466
But we did the first step anyway.

23
00:01:05,466 --> 00:01:08,900
And then the second step
is to take all the subsets in transaction

24
00:01:09,133 --> 00:01:13,100
having higher support
than the minimum 0.3 percent support.

25
00:01:13,633 --> 00:01:14,400
Okay.

26
00:01:14,400 --> 00:01:19,600
So this step two was actually completed
when we train the apriori on our data set.

27
00:01:19,600 --> 00:01:22,166
So the function did it itself.

28
00:01:22,166 --> 00:01:23,800
And that's the same for step three.

29
00:01:23,800 --> 00:01:27,033
The step three is to take
all the rules of these subsets

30
00:01:27,033 --> 00:01:30,033
that have a higher support
than 0.3 percent.

31
00:01:30,566 --> 00:01:33,233
And so we need to take
all the rules of these subsets

32
00:01:33,233 --> 00:01:36,933
that have higher confidence
than this minimum confidence of 40%.

33
00:01:37,200 --> 00:01:40,000
And so again
this was completed when we trained

34
00:01:40,000 --> 00:01:43,966
the apriori model on a data set
thanks to this apriori function.

35
00:01:44,366 --> 00:01:45,900
Okay. So step three done.

36
00:01:45,900 --> 00:01:48,200
And now
we're finally getting to step four.

37
00:01:48,200 --> 00:01:51,566
And that's the step that will lead us
to the visualization of the results.

38
00:01:51,966 --> 00:01:56,500
So this step four is to sort the rules
by their decreasing lift.

39
00:01:57,033 --> 00:01:59,533
As a rule
explaining the intuition tutorial

40
00:01:59,533 --> 00:02:03,600
to lift is the best metric
to measure the relevance of a rule.

41
00:02:03,800 --> 00:02:06,366
So that's
why we're sorting the rules by their lift

42
00:02:06,366 --> 00:02:09,366
rather than their confidence
or their support.

43
00:02:09,566 --> 00:02:12,833
So that's what we're going to do
in this single line of code.

44
00:02:13,300 --> 00:02:16,500
And to do this
we're going to use the inspect function.

45
00:02:16,766 --> 00:02:18,066
So we type here inspect.

46
00:02:19,333 --> 00:02:21,033
And so in this inspect function

47
00:02:21,033 --> 00:02:25,200
we can directly have a look at the rules
by typing rules here.

48
00:02:25,566 --> 00:02:28,100
And then you know if we want to have
a look at the first ten rules

49
00:02:28,100 --> 00:02:32,733
we need to specify here in brackets
one colon ten.

50
00:02:32,733 --> 00:02:35,100
So here we will get the first ten rules.

51
00:02:35,100 --> 00:02:38,900
But that's not very interesting
because this will just give us

52
00:02:38,900 --> 00:02:41,933
the ten first rules
found by our apriori model.

53
00:02:41,933 --> 00:02:44,466
So that's not going to be the rules.

54
00:02:44,466 --> 00:02:46,066
Having the highest lifts.

55
00:02:46,066 --> 00:02:48,100
That is the most relevant rules.

56
00:02:48,100 --> 00:02:51,900
So that's why right now
we need to sort the rules

57
00:02:52,066 --> 00:02:55,633
to get indeed the ten first rules
that have the ten highest lifts.

58
00:02:56,100 --> 00:02:57,400
So let's do this.

59
00:02:57,400 --> 00:03:01,033
We need to add another function here,
which is the sort function

60
00:03:01,033 --> 00:03:05,966
which obviously is used to sort any table
by a decreasing or increasing variable.

61
00:03:06,133 --> 00:03:10,266
So actually the first argument
of this sort function is rules

62
00:03:10,466 --> 00:03:13,733
that contains all the rules
found by our apriori model.

63
00:03:14,233 --> 00:03:17,400
And then the second argument is by

64
00:03:17,600 --> 00:03:20,966
which tells
by what we want to sort the rules.

65
00:03:21,200 --> 00:03:24,200
And of course
we want to sort the rules by their lift.

66
00:03:24,500 --> 00:03:28,566
So here we'll add by equals and then lift.

67
00:03:29,033 --> 00:03:32,766
And we keep this one content here
to get the ten first rules

68
00:03:32,766 --> 00:03:34,766
that have the ten highest lifts.

69
00:03:34,766 --> 00:03:35,100
All right.

70
00:03:35,100 --> 00:03:36,900
So that's actually ready.

71
00:03:36,900 --> 00:03:38,633
We are ready to visualize the result.

72
00:03:38,633 --> 00:03:43,500
That is to clearly see what the rules are
what the strongest rules are okay.

73
00:03:43,500 --> 00:03:46,500
So let's execute this.

74
00:03:47,000 --> 00:03:49,033
And here we go. All right.

75
00:03:49,033 --> 00:03:50,033
That's the rules.

76
00:03:50,033 --> 00:03:53,466
As I told you
we get very explicit and clear rules.

77
00:03:53,833 --> 00:03:57,300
The first rule is if people buy
mineral water and holy pasta,

78
00:03:57,300 --> 00:04:00,900
they will also buy olive oil
in 40% of the cases.

79
00:04:01,300 --> 00:04:04,433
Then the second rule is
if people buy spaghetti in tomato sauce,

80
00:04:04,633 --> 00:04:08,166
they will also buy ground beef
and 48% of the cases.

81
00:04:08,600 --> 00:04:11,900
Then if people buy French fries
and herb and pepper, they will also buy

82
00:04:11,900 --> 00:04:14,900
ground beef and 46% of the cases.

83
00:04:15,000 --> 00:04:16,466
So that's how the rules work.

84
00:04:16,466 --> 00:04:19,200
That's how they are presented
thanks to this package.

85
00:04:19,200 --> 00:04:23,033
And so these are the first ten rules
with the ten highest lifts.

86
00:04:23,466 --> 00:04:27,366
However do you see something that is not
necessarily very relevant here.

87
00:04:27,933 --> 00:04:30,600
Well if we have a better
look at these rules we can notice

88
00:04:30,600 --> 00:04:34,566
that some products are present
in a set of items

89
00:04:34,733 --> 00:04:39,033
not because they make a good association,
but because they have a high support.

90
00:04:39,500 --> 00:04:42,400
A good example of
this is the chocolate here.

91
00:04:42,400 --> 00:04:46,066
This set of items, composed of chocolate
and herb and pepper, corresponds

92
00:04:46,066 --> 00:04:49,500
to some customers that bought chocolate
and herb and pepper in their basket.

93
00:04:49,866 --> 00:04:52,500
And according to this rule, well,
these customers

94
00:04:52,500 --> 00:04:55,666
also buy ground beef in 44% of the cases.

95
00:04:56,066 --> 00:04:57,966
But the thing is,
they don't want to buy ground beef

96
00:04:57,966 --> 00:05:00,966
because they had some chocolate
in their basket in the first place.

97
00:05:01,133 --> 00:05:05,033
Again, that doesn't make any sense, but
it's not because the confidence is too low

98
00:05:05,033 --> 00:05:08,400
this time, it's because the chocolate
has a very high support.

99
00:05:08,833 --> 00:05:09,800
We can see that here.

100
00:05:09,800 --> 00:05:13,766
Chocolate is the fifth product
most purchased by these French

101
00:05:13,766 --> 00:05:15,500
customers in the south of France,

102
00:05:15,500 --> 00:05:19,400
and therefore this product, chocolate here
falls in a lot of baskets.

103
00:05:19,633 --> 00:05:23,766
And especially in this basket number
six here, composed of chocolate and herb

104
00:05:23,766 --> 00:05:24,433
and pepper,

105
00:05:24,433 --> 00:05:27,333
as well as this basket number seven here
composed of chocolate,

106
00:05:27,333 --> 00:05:28,900
mineral water and shrimp.

107
00:05:28,900 --> 00:05:32,366
And by the way, in this same basket
there is mineral water

108
00:05:32,500 --> 00:05:35,433
that also falls in this eight basket here.

109
00:05:35,433 --> 00:05:39,166
And that's because mineral water is
the most purchased product in the store.

110
00:05:39,333 --> 00:05:42,000
So of course
it's falling in a lot of baskets.

111
00:05:42,000 --> 00:05:44,433
So these products have high support.

112
00:05:44,433 --> 00:05:47,433
However, this doesn't mean
that we have to change the support now

113
00:05:47,533 --> 00:05:51,566
because first we still want to validate
our first business points, which was to

114
00:05:51,600 --> 00:05:55,200
consider only the products that are bought
at least 3 or 4 times a day.

115
00:05:55,466 --> 00:05:58,600
But what might be helpful to change
now is the confidence.

116
00:05:58,966 --> 00:06:03,100
Because indeed, we asked for the rules
that have at least a 40% confidence.

117
00:06:03,100 --> 00:06:07,100
You can see that all the rules here
have confidence over 40%.

118
00:06:07,433 --> 00:06:09,866
And the reason the rules have high
confidence

119
00:06:09,866 --> 00:06:12,866
is because the rules are associated
to the baskets

120
00:06:12,900 --> 00:06:16,200
that contain
the most purchased products in the store.

121
00:06:16,633 --> 00:06:20,100
Because of course, if the baskets contain
the most purchased products in the store,

122
00:06:20,400 --> 00:06:24,700
well, these products will go together
and the basket, and it won't be

123
00:06:24,700 --> 00:06:28,433
because of some association rules
related to this principle.

124
00:06:28,433 --> 00:06:29,500
People who bought also.

125
00:06:29,500 --> 00:06:31,033
But but it will be because

126
00:06:31,033 --> 00:06:34,600
simply the baskets contain
the most purchased products overall.

127
00:06:34,966 --> 00:06:38,533
So in order to avoid this, the first idea
would be to change the supports.

128
00:06:38,733 --> 00:06:42,400
But we don't want to change our business
related starting points, you know,

129
00:06:42,400 --> 00:06:45,400
to consider the products, purchase
at least 3 or 4 times a day.

130
00:06:45,533 --> 00:06:49,600
So the other idea that remains left is,
of course, to change to confidence.

131
00:06:49,933 --> 00:06:51,533
Because if we reduce the confidence

132
00:06:51,533 --> 00:06:53,833
now, well,
we won't get these specific rules

133
00:06:53,833 --> 00:06:55,200
that exist because they're

134
00:06:55,200 --> 00:06:58,800
associated to the most purchased products
that fall in the same baskets.

135
00:06:59,133 --> 00:07:02,133
But we will get the most relevant rules
that we are looking for

136
00:07:02,133 --> 00:07:04,100
and that are related to this principle.

137
00:07:04,100 --> 00:07:05,866
People who bought also bought.

138
00:07:05,866 --> 00:07:07,900
So that's
what we're going to do right now.

139
00:07:07,900 --> 00:07:10,600
We're going to change
the confidence will reduce it.

140
00:07:10,600 --> 00:07:13,600
And what we're going to do
is what we already did the first time.

141
00:07:13,733 --> 00:07:16,500
You know, remember
we had a confidence of 0.8.

142
00:07:16,500 --> 00:07:17,933
That gave us no rule.

143
00:07:17,933 --> 00:07:22,200
And then we divided it by two
to obtain this confidence of 0.4.

144
00:07:22,333 --> 00:07:25,333
That gave us these rules
related to the most purchased products.

145
00:07:25,566 --> 00:07:28,800
So now we're going to do the same
and divide the confidence by two.

146
00:07:28,800 --> 00:07:31,866
Again, to get a confidence of 20%.

147
00:07:32,200 --> 00:07:35,200
And that should lead us
to some more relevant rules

148
00:07:35,400 --> 00:07:39,666
that are related to the association rules
that we're looking for,

149
00:07:39,866 --> 00:07:41,300
related to the principle.

150
00:07:41,300 --> 00:07:43,200
People who bought also bought.

151
00:07:43,200 --> 00:07:45,466
Okay, so let's do it.

152
00:07:45,466 --> 00:07:48,000
Let's try these new rules.

153
00:07:48,000 --> 00:07:51,000
Because by selecting this line
and executing it,

154
00:07:51,033 --> 00:07:53,300
the apriori model will be trained again.

155
00:07:53,300 --> 00:07:54,900
And so we will get some new rules.

156
00:07:54,900 --> 00:07:57,300
So let's do it. Execute.

157
00:07:57,300 --> 00:08:01,466
And here it is
okay first of all we get a lot more rules.

158
00:08:01,466 --> 00:08:04,366
We get 1348 rules.

159
00:08:04,366 --> 00:08:07,666
And that's expected of course
because since we reduced

160
00:08:07,666 --> 00:08:13,000
the confidence down to 0.2, well of course
the algorithm found a lot more rules.

161
00:08:13,000 --> 00:08:17,133
But don't worry,
we will not have a look at the 1348 rules.

162
00:08:17,300 --> 00:08:21,533
We will still look at the ten first rules
with the ten highest lifts, and that's

163
00:08:21,533 --> 00:08:25,500
what we're going to do right now
by selecting this line and execute it.

164
00:08:25,966 --> 00:08:28,133
And here are the new rules okay.

165
00:08:28,133 --> 00:08:29,166
So let's have a look.

166
00:08:29,166 --> 00:08:33,333
The first rule is if people buy
mineral water and holy pasta

167
00:08:33,333 --> 00:08:36,600
they will buy olive oil
in 40% of the cases.

168
00:08:36,933 --> 00:08:37,533
Okay.

169
00:08:37,533 --> 00:08:38,933
So that's the rule

170
00:08:38,933 --> 00:08:41,933
that makes a lot of sense
even if there is still mineral water.

171
00:08:42,166 --> 00:08:43,500
But, you know,

172
00:08:43,500 --> 00:08:44,200
it's still a rule

173
00:08:44,200 --> 00:08:48,133
that makes sense because, you know,
this might be related to some people who

174
00:08:48,333 --> 00:08:53,333
are looking to have some healthy meals
with mineral water, whole wheat pasta,

175
00:08:53,500 --> 00:08:56,433
and of course, olive
oil is very healthy as well.

176
00:08:56,433 --> 00:09:00,766
So these go well together and actually
this is kind of a relevant rule.

177
00:09:01,100 --> 00:09:04,100
Olive oil should be placed
not too far from whole wheat pasta.

178
00:09:04,900 --> 00:09:05,166
Okay.

179
00:09:05,166 --> 00:09:08,400
Then the second rule is
if people buy frozen vegetables,

180
00:09:08,400 --> 00:09:11,933
milk, mineral water,
then they will buy soup.

181
00:09:12,200 --> 00:09:15,200
And 27% of the cases.

182
00:09:15,233 --> 00:09:18,033
Again,
this is a rule that actually makes sense.

183
00:09:18,033 --> 00:09:19,500
Still, related to the need

184
00:09:19,500 --> 00:09:23,200
of having healthy meals
and milk can go quite well with soup.

185
00:09:23,500 --> 00:09:26,500
Well,
I know this is the case for French people.

186
00:09:26,700 --> 00:09:29,700
French people do
tend to put some milk in their soup.

187
00:09:30,033 --> 00:09:33,033
And oh, by the way,
speaking of a French tradition,

188
00:09:33,200 --> 00:09:36,233
this is typically what French people love.

189
00:09:36,233 --> 00:09:38,733
Absolutely. From us. Blown with honey.

190
00:09:38,733 --> 00:09:40,833
So, for example, for those of you
who don't know, it's

191
00:09:40,833 --> 00:09:43,900
kind of some good cheese,
I invite you to look at Wikipedia.

192
00:09:44,166 --> 00:09:47,166
But anyway, this goes
very well with honey.

193
00:09:47,366 --> 00:09:51,300
You know, in a lot of French restaurants
you will find for marshmallow with honey

194
00:09:51,300 --> 00:09:52,700
in the dessert menu.

195
00:09:52,700 --> 00:09:56,366
But, people also make this at their place.

196
00:09:56,366 --> 00:09:59,700
So that's kind of a very good association
rule,

197
00:10:00,033 --> 00:10:02,566
even if for marshmallow is very different
than honey.

198
00:10:02,566 --> 00:10:05,433
Well, these two products associate
very well.

199
00:10:05,433 --> 00:10:09,533
And, you know, the direction counts here,
because if we want to buy from us below,

200
00:10:09,566 --> 00:10:10,866
then we want to buy honey.

201
00:10:10,866 --> 00:10:13,866
Because honey goes
very well onto from us below.

202
00:10:14,000 --> 00:10:17,866
But if we buy honey, we don't necessarily
want to buy from us below

203
00:10:17,900 --> 00:10:20,900
because we don't want to have from us,
not in our honey.

204
00:10:21,066 --> 00:10:26,100
It's rather we want to have honey on our
from us and not from us, but on our honey.

205
00:10:26,566 --> 00:10:27,600
All right.

206
00:10:27,600 --> 00:10:29,200
And then what do we have? We have.

207
00:10:29,200 --> 00:10:33,466
If people buy spaghetti and tomato sauce,
they want to buy ground beef.

208
00:10:33,566 --> 00:10:35,266
Well, that makes a lot of sense.

209
00:10:35,266 --> 00:10:38,566
That's of course,
to make some spaghetti bolognese.

210
00:10:39,000 --> 00:10:41,400
All right. So interesting. Quite classic.

211
00:10:41,400 --> 00:10:42,766
You know, we don't need

212
00:10:42,766 --> 00:10:46,366
a, machine learning algorithm
to find out about this rule, but,

213
00:10:46,666 --> 00:10:51,800
you know, it's actually what French
people love associating in their basket.

214
00:10:51,800 --> 00:10:55,600
So, of course, the ground beef shouldn't
be too far from the spaghetti

215
00:10:55,766 --> 00:10:57,600
and the tomato sauce.

216
00:10:57,600 --> 00:11:00,600
And then this rule,
if people buy light cream,

217
00:11:00,866 --> 00:11:03,866
they will buy chicken in 29% of the cases.

218
00:11:04,366 --> 00:11:07,366
So not necessarily obvious.

219
00:11:07,500 --> 00:11:08,566
You know,

220
00:11:08,566 --> 00:11:12,333
if the manager of the store
had to place the product itself

221
00:11:12,333 --> 00:11:15,633
without any algorithm,
this manager would not have necessarily

222
00:11:15,633 --> 00:11:19,133
thought to place the chicken
next to the light cream.

223
00:11:19,700 --> 00:11:23,466
And, you know, if we try to explain this
rule, it's probably because,

224
00:11:23,700 --> 00:11:27,700
you know, people who buy some light cream
want to pay attention to what they eat.

225
00:11:27,933 --> 00:11:32,866
And so since chicken is a lighter meat
and maybe healthier meat than red meat

226
00:11:32,933 --> 00:11:36,433
like ground beef, well, if they buy
light cream and want to associate it

227
00:11:36,433 --> 00:11:39,433
with some meat, well,
they would rather choose chicken.

228
00:11:39,766 --> 00:11:43,966
That's you know, if we try to explain
the rules with some good sense

229
00:11:44,566 --> 00:11:45,366
and then what do we have?

230
00:11:45,366 --> 00:11:49,766
We have if people buy pasta,
they will buy scallop in 37% of the cases.

231
00:11:50,166 --> 00:11:51,966
Well why not? That goes well together.

232
00:11:51,966 --> 00:11:53,566
Well, that's simply what

233
00:11:53,566 --> 00:11:57,000
French people in the south of France
love to eat in their meals.

234
00:11:57,400 --> 00:11:59,733
And maybe that's also related
to French taste

235
00:11:59,733 --> 00:12:02,733
that French people associate
like cream with chicken.

236
00:12:02,833 --> 00:12:05,833
If we were in India,
it would probably have been butter here

237
00:12:06,066 --> 00:12:09,066
for, you know, butter chicken,
which is a very good Indian meal.

238
00:12:09,366 --> 00:12:09,666
Okay.

239
00:12:09,666 --> 00:12:12,566
And then we have French fries
and herb and pepper that go

240
00:12:12,566 --> 00:12:13,966
very well with ground beef.

241
00:12:13,966 --> 00:12:16,600
Of course, that's a classic French meal.

242
00:12:16,600 --> 00:12:18,866
then cereal spaghetti, ground beef.

243
00:12:18,866 --> 00:12:19,200
Okay.

244
00:12:19,200 --> 00:12:21,733
That's a rule
that doesn't make necessarily much sense.

245
00:12:21,733 --> 00:12:25,500
And maybe it's due to the same logic that
I was explaining earlier that is that,

246
00:12:25,500 --> 00:12:29,266
you know, a lot of people buy cereals,
that a lot of people buy spaghetti.

247
00:12:29,466 --> 00:12:32,466
So spaghetti and cereals
often fall in the same basket.

248
00:12:32,833 --> 00:12:36,466
And since a lot of people associate
spaghetti with ground beef, well,

249
00:12:36,466 --> 00:12:40,466
we find this rule if people buy cereals
and spaghetti, they will buy ground beef.

250
00:12:40,533 --> 00:12:42,600
So let's be careful with this one.

251
00:12:42,600 --> 00:12:45,933
Cereals is not necessarily
associated to ground beef.

252
00:12:46,200 --> 00:12:49,900
But you know
we can investigate further okay.

253
00:12:50,233 --> 00:12:52,166
And then we have these last two rules.

254
00:12:52,166 --> 00:12:55,166
If people buy frozen vegetables
mineral water and soup

255
00:12:55,200 --> 00:12:58,500
they're likely to buy
milk in 60% of the cases.

256
00:12:58,500 --> 00:13:01,200
Well, this rule is probably related
to two facts.

257
00:13:01,200 --> 00:13:04,766
First, the fact that milk goes very well
with soup for French people,

258
00:13:05,066 --> 00:13:08,433
and also because, you know,
all this looks very healthy.

259
00:13:08,666 --> 00:13:10,866
So that goes quite well together.

260
00:13:10,866 --> 00:13:14,500
And then the rest rules
French fries and ground beef that go again

261
00:13:14,500 --> 00:13:16,033
very well with herb and pepper. Yes.

262
00:13:16,033 --> 00:13:17,066
Because of course

263
00:13:17,066 --> 00:13:20,700
here we have French fries and herb
and pepper that lead to ground beef.

264
00:13:21,000 --> 00:13:25,433
Well of course an association rules
the other direction can sometimes be true.

265
00:13:25,433 --> 00:13:27,166
And that's exactly the case here.

266
00:13:27,166 --> 00:13:28,100
Here we have

267
00:13:28,100 --> 00:13:32,233
if people buy ground beef and French fries
they will also buy herb and pepper.

268
00:13:32,233 --> 00:13:36,300
That's kind of a triangle association
rules here that we observe with the

269
00:13:36,300 --> 00:13:40,800
three sides of the triangle being French
fries, ground beef and herb and pepper.

270
00:13:41,366 --> 00:13:42,600
So that's quite common,

271
00:13:42,600 --> 00:13:46,866
but it's not always the case
to observe the two directions okay.

272
00:13:47,133 --> 00:13:50,600
So that's definitely very helpful
for this tour.

273
00:13:50,600 --> 00:13:54,300
Now thanks to these rules
we can experience a new placement

274
00:13:54,300 --> 00:13:55,300
of the products here.

275
00:13:55,300 --> 00:13:59,133
We can also look at the 21st rules
with the 21st lifts.

276
00:13:59,600 --> 00:14:02,466
We won't do it now
because we get the point here.

277
00:14:02,466 --> 00:14:07,066
However, what we can do is you know,
we actually try another support.

278
00:14:07,233 --> 00:14:11,166
We tried three values for the confidence,
but we only tried one

279
00:14:11,166 --> 00:14:12,600
value of the support.

280
00:14:12,600 --> 00:14:15,100
And remember we had this hypothesis.

281
00:14:15,100 --> 00:14:17,033
I'm in this business starting point.

282
00:14:17,033 --> 00:14:18,133
When we said that we wanted

283
00:14:18,133 --> 00:14:21,666
to consider the products that are
but at least 3 or 4 times a day.

284
00:14:21,900 --> 00:14:27,000
Well, this support here, 0.03,
is related to the business starting point

285
00:14:27,000 --> 00:14:30,000
of considering products that are bought
at least three times a day.

286
00:14:30,400 --> 00:14:33,866
Then what if we consider now products
that are bought at least four times a day?

287
00:14:34,200 --> 00:14:36,166
So that's what
we're going to do right now.

288
00:14:36,166 --> 00:14:39,200
This is the only other support value
that we're going to try.

289
00:14:39,500 --> 00:14:40,866
So let's try this.

290
00:14:40,866 --> 00:14:43,833
Remember
the computation that leads to the support.

291
00:14:43,833 --> 00:14:46,833
If we come to the products that are bought
at least four times a day,

292
00:14:47,100 --> 00:14:50,100
then on average
these products will be bought

293
00:14:50,400 --> 00:14:53,400
four times seven times a week.

294
00:14:53,666 --> 00:14:56,400
And then in order to get the support,
we need to divided

295
00:14:56,400 --> 00:15:01,466
by the total number of transactions,
which is 7005 hundreds.

296
00:15:02,266 --> 00:15:04,033
All right. So let's see what we get.

297
00:15:04,033 --> 00:15:06,300
We get a minimum support of

298
00:15:07,333 --> 00:15:09,666
0.00 37.

299
00:15:09,666 --> 00:15:14,266
So this time if we round it
we get a minimum support of 0.04.

300
00:15:14,266 --> 00:15:18,400
So let's try
this 10.004 that corresponds to products

301
00:15:18,666 --> 00:15:22,233
that were purchased at least four times
a day instead of three previously.

302
00:15:22,500 --> 00:15:26,900
So let's train the apriori model again
to obtain the new rules.

303
00:15:27,133 --> 00:15:30,700
So I'm going to execute this
and we get the new rules.

304
00:15:30,900 --> 00:15:34,866
And since we increase the support
we find a smaller

305
00:15:34,866 --> 00:15:38,533
number of rules
811 rules compared to previously.

306
00:15:38,533 --> 00:15:40,933
You know, we had more than 1000 rules.

307
00:15:40,933 --> 00:15:45,600
That's because we increased the support
and kept the same 20% confidence.

308
00:15:45,900 --> 00:15:46,500
Okay.

309
00:15:46,500 --> 00:15:50,566
So now let's have a look at these
new rules ordered by their decreasing lift

310
00:15:51,166 --> 00:15:53,733
okay. So let's execute this.

311
00:15:53,733 --> 00:15:56,100
And here are the new rules
okay. Very interesting.

312
00:15:56,100 --> 00:16:00,766
First by increasing the supports
and therefore excluding some products

313
00:16:00,766 --> 00:16:05,466
that have supports between 0.03 and 0.04.

314
00:16:05,866 --> 00:16:08,166
Well remember this first rule.

315
00:16:08,166 --> 00:16:11,000
If people buy light cream
they will also buy chicken.

316
00:16:11,000 --> 00:16:13,400
Well this rule now became the top rule.

317
00:16:13,400 --> 00:16:15,733
That's
definitely a great rule to consider.

318
00:16:15,733 --> 00:16:18,833
And chicken
should definitely be close to light cream.

319
00:16:19,300 --> 00:16:19,566
Okay.

320
00:16:19,566 --> 00:16:22,566
And then we have if people buy pasta
they buy scallop.

321
00:16:22,700 --> 00:16:24,933
Well we also had this rule before.

322
00:16:24,933 --> 00:16:26,600
However we have a new rule here.

323
00:16:26,600 --> 00:16:29,600
If people buy pasta
they will also buy shrimp.

324
00:16:29,700 --> 00:16:31,800
Yes of course this is the south of France.

325
00:16:31,800 --> 00:16:33,300
People are close to the sea.

326
00:16:33,300 --> 00:16:36,866
Well, either the Mediterranean Sea
or the Atlantic Ocean.

327
00:16:37,166 --> 00:16:41,200
And of course, people love
associating shrimp with their pasta.

328
00:16:41,666 --> 00:16:44,833
That's a very common
and exquisite meal in the south of France.

329
00:16:45,433 --> 00:16:46,166
And then what do we have?

330
00:16:46,166 --> 00:16:49,166
We have if people buy
eggs and ground beef,

331
00:16:49,433 --> 00:16:53,033
they're likely to buy herb and pepper
in 20% of the cases.

332
00:16:53,333 --> 00:16:57,200
Well, of course, having pepper
is very good on both eggs and ground beef.

333
00:16:57,733 --> 00:17:02,400
And then we have the same kind of rules
as before, except for this new rule here.

334
00:17:02,833 --> 00:17:07,500
French people love to associate
mushroom cream sauce with scallop.

335
00:17:07,833 --> 00:17:09,600
So indeed it's very good meal.

336
00:17:09,600 --> 00:17:12,600
Mushroom cream sauce goes
very well with scallop.

337
00:17:12,666 --> 00:17:13,033
Okay.

338
00:17:13,033 --> 00:17:17,766
So interesting new rules that we obtain
when we increase the support.

339
00:17:18,133 --> 00:17:22,100
So the manager of the store
now should, you know, experience

340
00:17:22,100 --> 00:17:26,133
this new rules that we observe here
by placing the new products of these rules

341
00:17:26,133 --> 00:17:27,233
next to each other,

342
00:17:27,233 --> 00:17:31,000
then experience for a few weeks
and then observe the impact on the sales,

343
00:17:31,266 --> 00:17:34,266
how much the sales increase,
how much the revenue increases,

344
00:17:34,466 --> 00:17:36,633
and then see
if the business goals are achieved.

345
00:17:36,633 --> 00:17:39,600
And if that's the case,
try to strengthen the rules

346
00:17:39,600 --> 00:17:43,900
or try some more powerful rules
by changing the support in confidence or

347
00:17:43,900 --> 00:17:47,200
on the other hand, if the business goals
are not achieved, well, same.

348
00:17:47,200 --> 00:17:50,133
We can try to get some new rules
by increasing the confidence

349
00:17:50,133 --> 00:17:51,700
and maybe the support.

350
00:17:51,700 --> 00:17:54,133
Well, that's all related to experience.

351
00:17:54,133 --> 00:17:57,900
That's what data
analysts and managers do in retail stores,

352
00:17:57,900 --> 00:18:01,800
whether there are online stores
or grocery stores or any kind of stores.

353
00:18:02,000 --> 00:18:05,333
Well, they use these association rules
and update them.

354
00:18:05,333 --> 00:18:09,166
And of course, combine these rules
with other recommendation system

355
00:18:09,166 --> 00:18:13,133
techniques like collaborative filtering
with, you know, the user profiles

356
00:18:13,133 --> 00:18:17,700
that can add some additional relevant
info, and also other more advanced

357
00:18:17,700 --> 00:18:21,433
techniques like the neighborhood model,
the latent factor models.

358
00:18:21,433 --> 00:18:25,133
Well, they combine a lot of models
to increase the sales and the revenue.

359
00:18:25,800 --> 00:18:28,866
but here you have a very good technique,
very powerful technique,

360
00:18:28,866 --> 00:18:30,233
association rules.

361
00:18:30,233 --> 00:18:35,000
So congratulations for having implemented
this first recommendation system

362
00:18:35,000 --> 00:18:36,000
in some way.

363
00:18:36,000 --> 00:18:40,033
And I really hope this will be useful
for your business and your work.

364
00:18:40,233 --> 00:18:42,733
And don't hesitate
if you have any questions about this.

365
00:18:42,733 --> 00:18:45,300
We will answer your questions
very quickly.

366
00:18:45,300 --> 00:18:47,333
So that was the apriori algorithm.

367
00:18:47,333 --> 00:18:49,333
I was very happy to build this model
with you.

368
00:18:49,333 --> 00:18:50,733
Congratulations again.

369
00:18:50,733 --> 00:18:54,766
Now in the next section we are going
to implement the Eclat algorithm,

370
00:18:54,766 --> 00:18:58,866
which is very close to the apriori
algorithm but much more simple.

371
00:18:58,866 --> 00:19:01,866
So that's another very good solution
if you want to be very efficient

372
00:19:01,933 --> 00:19:05,100
and don't have too much time to experience
different values of the support

373
00:19:05,333 --> 00:19:08,033
and the confidence, because indeed
you will see that

374
00:19:08,033 --> 00:19:11,200
there is no confidence
parameter in the algorithm.

375
00:19:11,666 --> 00:19:12,133
So I look

376
00:19:12,133 --> 00:19:15,800
forward to building this new association
rule model with you in the next section.

377
00:19:15,933 --> 00:19:17,633
Until then, enjoy machine learning.