﻿1
00:00:01,733 --> 00:00:04,133
Hello and welcome back to the course
on Machine Learning.

2
00:00:04,133 --> 00:00:07,900
Today we're continuing
exploring association rule learning.

3
00:00:07,900 --> 00:00:10,900
And we're talking about the intuition
behind the model.

4
00:00:11,200 --> 00:00:15,100
So the cloud model is very, very simple.

5
00:00:15,100 --> 00:00:17,066
After we have already studied

6
00:00:17,066 --> 00:00:20,900
the apriori model, it's
kind of like a simplified, version.

7
00:00:21,000 --> 00:00:23,166
All right. So let's have a look.

8
00:00:23,166 --> 00:00:26,266
it also talks about the cloud
models, talks about people

9
00:00:26,266 --> 00:00:27,300
who are bored, also bored.

10
00:00:27,300 --> 00:00:30,300
So it's kind of like a,
recommender system.

11
00:00:30,366 --> 00:00:33,566
And similar to what we had in the,

12
00:00:34,000 --> 00:00:37,000
a, a algorithm.

13
00:00:37,266 --> 00:00:40,500
here we've got, for instance, movies
and we've got some potential rules.

14
00:00:40,500 --> 00:00:42,600
So basically exactly the same things.

15
00:00:42,600 --> 00:00:45,666
If you've got, your,

16
00:00:45,666 --> 00:00:49,733
movie lists or people,
the movies that people liked, people

17
00:00:49,733 --> 00:00:53,100
who like movie one hour
or just generally looks like

18
00:00:53,700 --> 00:00:57,133
if somebody likes movie one,
they're very likely to like movie two,

19
00:00:58,000 --> 00:01:01,000
if they like movie two,
they're likely to like movie four.

20
00:01:01,100 --> 00:01:03,466
If they like movie one,
they might like movie three.

21
00:01:03,466 --> 00:01:08,066
Again, these rules have
will have different, strength.

22
00:01:08,066 --> 00:01:12,000
But here we are actually going
to be talking about rules per se,

23
00:01:12,000 --> 00:01:15,366
because the cloud model is different
to the apriori model.

24
00:01:15,366 --> 00:01:20,833
in the parallel model we came up
with rules towards the end that's output.

25
00:01:20,833 --> 00:01:23,833
And we based on the left, we could, judge,

26
00:01:23,866 --> 00:01:27,133
the strength of each rule, whereas here

27
00:01:27,566 --> 00:01:31,633
we are, going to be talking about sets
and you'll see why just now.

28
00:01:31,833 --> 00:01:35,366
So here we've got the market basket
optimization.

29
00:01:35,366 --> 00:01:40,066
Same thing that people who buy burgers
are likely to buy French fries as well.

30
00:01:40,066 --> 00:01:42,466
People who I buy vegetables,
I like to buy fruits.

31
00:01:42,466 --> 00:01:44,200
And these are just some potential rules.

32
00:01:44,200 --> 00:01:46,200
So we're not saying they're, strong

33
00:01:47,200 --> 00:01:49,900
or some
potential outcomes that we're looking at.

34
00:01:49,900 --> 00:01:52,000
We're not saying that they're strong
or we're not selecting

35
00:01:52,000 --> 00:01:54,533
and we're just saying
what could potentially be.

36
00:01:54,533 --> 00:01:58,700
And then the cloud model is responsible
for actually

37
00:01:58,700 --> 00:02:03,300
going through all of these combinations
and telling us what we should focus on.

38
00:02:04,633 --> 00:02:04,933
all right.

39
00:02:04,933 --> 00:02:08,833
So in the cloud model,
just like in the a priori model,

40
00:02:08,833 --> 00:02:12,166
we have the support factor.

41
00:02:12,166 --> 00:02:17,666
So they're producing the apriori model
we had or the algorithm we had support.

42
00:02:17,666 --> 00:02:20,400
We had confidence
and we had lift in their cloud model.

43
00:02:20,400 --> 00:02:21,900
We only have support.

44
00:02:21,900 --> 00:02:25,333
So we only are looking at okay,

45
00:02:25,333 --> 00:02:29,633
so people who are watching, a certain,

46
00:02:30,266 --> 00:02:33,700
certain combinations of movies,
how often does this happen?

47
00:02:33,700 --> 00:02:37,366
And here just bear in mind
that, doesn't mean just one movie.

48
00:02:37,366 --> 00:02:39,800
And this was the same for a priori.

49
00:02:39,800 --> 00:02:40,500
It was just easier

50
00:02:40,500 --> 00:02:45,600
for us to understand the intuition
based on one movie or one product.

51
00:02:45,600 --> 00:02:51,766
But actually, and I what they stand for
is a set of items or a set of movies.

52
00:02:51,766 --> 00:02:56,000
So as specifically in the cloud model,
it's doesn't really make sense to,

53
00:02:56,766 --> 00:03:01,200
look at, you know, an item by itself
because we don't have the,

54
00:03:01,600 --> 00:03:05,400
confidence and the lift factors.

55
00:03:05,400 --> 00:03:06,433
We're only looking at support.

56
00:03:06,433 --> 00:03:10,733
So we're just looking at how frequently
does this set of items, occur.

57
00:03:10,866 --> 00:03:11,566
So if we just going to

58
00:03:11,566 --> 00:03:15,466
look at a set of items of, of items
which contains only one item,

59
00:03:16,033 --> 00:03:18,633
then we just looking at the frequency
and how what

60
00:03:18,633 --> 00:03:21,033
what is the popularity of movies.
And that is very trivial.

61
00:03:21,033 --> 00:03:22,200
So we're not going to be looking at that.

62
00:03:22,200 --> 00:03:25,933
We're going to be aiming
for at least two items in a set.

63
00:03:26,066 --> 00:03:31,266
And therefore M here
stands for a set of two movies or more.

64
00:03:31,633 --> 00:03:37,166
And what we're calculating for a support,
we're calculating.

65
00:03:37,166 --> 00:03:38,000
Okay.

66
00:03:38,000 --> 00:03:42,666
what is how often does this
set of two movies, let's say,

67
00:03:43,033 --> 00:03:48,433
interstellar or and Ex Machina,
how often does it

68
00:03:48,433 --> 00:03:52,433
occur in all of the watch lists,
or what percentage of watch lists

69
00:03:52,433 --> 00:03:56,066
or what percentage of,

70
00:03:56,066 --> 00:04:00,366
lists of movies that people liked contain
those two together?

71
00:04:00,400 --> 00:04:02,566
Not just one of them,
but those two together.

72
00:04:02,566 --> 00:04:06,500
And let's say if, if hypothetically,

73
00:04:06,500 --> 00:04:11,633
if 100% of the lists that you had
in a large data set contained both movies

74
00:04:11,633 --> 00:04:16,333
together, then that would imply that,
you know, anybody who likes

75
00:04:16,566 --> 00:04:20,200
interstellar, likes Ex Machina, like Ex
Machina likes interstellar,

76
00:04:20,466 --> 00:04:24,266
and pretty much so if anybody has seen
even one of those movies

77
00:04:24,266 --> 00:04:27,266
that you need to recommend that movie
or the other one to them,

78
00:04:27,633 --> 00:04:30,933
if you or if you had like 80% of

79
00:04:30,933 --> 00:04:34,666
the list of your lists
had those two movies together,

80
00:04:34,900 --> 00:04:38,600
that basically mean there's a high
likelihood that they come in pairs, right?

81
00:04:38,600 --> 00:04:39,866
That if somebody liked one of them,

82
00:04:39,866 --> 00:04:42,166
then they'll like that
same thing for transaction.

83
00:04:42,166 --> 00:04:43,433
Like if you have,

84
00:04:43,433 --> 00:04:48,700
chips and burgers and, you know, 75%
of all of your orders, right?

85
00:04:48,700 --> 00:04:53,300
Then if somebody is just buying, burgers,
then they're likely to,

86
00:04:53,900 --> 00:04:58,200
then when you recommend chips to them,
there's a 75% chance that they will also

87
00:04:58,700 --> 00:05:01,733
be interested or will like to buy chips
with their burgers.

88
00:05:02,233 --> 00:05:06,000
And that's it's a very,
very trivial, approach.

89
00:05:06,833 --> 00:05:09,866
and that's, that's pretty much it.

90
00:05:09,866 --> 00:05:11,866
That's all there is to that model.

91
00:05:11,866 --> 00:05:12,900
It's much faster.

92
00:05:12,900 --> 00:05:16,333
And the steps involved are,
set a minimum support.

93
00:05:16,333 --> 00:05:21,166
So you want to set your, support level
at which you want to,

94
00:05:21,166 --> 00:05:24,866
only after below,
which you want to disregard anything.

95
00:05:25,466 --> 00:05:28,633
then you take all the subsets
and transactions having higher support

96
00:05:28,633 --> 00:05:32,300
and mind and support and then you sort
this subset in decreasing support.

97
00:05:32,300 --> 00:05:36,933
And basically at the top
you will have the most, the strongest,

98
00:05:37,300 --> 00:05:40,900
combinations of items,
which you should look at,

99
00:05:41,200 --> 00:05:44,666
maybe, you know, you'll look at the top
ten or to file or something like that.

100
00:05:44,666 --> 00:05:46,133
So that's pretty much it.

101
00:05:46,133 --> 00:05:48,200
That's all the A club model is.

102
00:05:48,200 --> 00:05:51,966
And as you can see, it's much easier
to understand after you already know,

103
00:05:52,266 --> 00:05:55,166
a bit about the apriori.

104
00:05:55,166 --> 00:05:57,000
All right. Hope you enjoyed this tutorial.

105
00:05:57,000 --> 00:06:00,000
And if we go to learned to

106
00:06:00,033 --> 00:06:02,833
implement this in practice
and I'll see you here next time.

107
00:06:02,833 --> 00:06:04,300
And till then happy analyzing.
