0
1
00:00:00,300 --> 00:00:05,040
To create our word cloud for our ham messages and our spam messages,
1

2
00:00:05,250 --> 00:00:07,650
we're gonna use two different masks.
2

3
00:00:07,680 --> 00:00:15,980
I've got thumbs up for the non-spam messages and I've got thumbs down for the spam messages.
3

4
00:00:16,080 --> 00:00:20,730
These are gonna be our masks for our respective word clouds.
4

5
00:00:20,730 --> 00:00:25,770
The dataset backing our word clouds will come from some of the previous work we've done. So we've got
5

6
00:00:25,800 --> 00:00:34,250
a flattened list of non-spam words and we've got a flattened list of spam words.
6

7
00:00:34,280 --> 00:00:37,970
This is the data that we're going to feed into our word clouds.
7

8
00:00:37,970 --> 00:00:40,910
So let's tackle our ham messages first.
8

9
00:00:40,910 --> 00:00:52,070
I'm going to add a markdown cell here and write "Word Cloud of Ham and Spam Messages".
9

10
00:00:52,070 --> 00:00:54,610
Next I'm going to add some constants.
10

11
00:00:54,710 --> 00:01:01,790
I'm going to add the thumbs up and thumbs down files, so I'll paste the relative path for the skull file
11

12
00:01:02,540 --> 00:01:12,650
and just rename these to "THUMBS_UP_FILE" and "THUMBS_DOWN_FILE".
12

13
00:01:12,650 --> 00:01:15,170
Of course the icon names have to change as well.
13

14
00:01:15,410 --> 00:01:28,240
So this one well call "thumbs-up.png", and this one we'll call "thumbs-down.
14

15
00:01:28,240 --> 00:01:31,700
png". Shift+Enter,
15

16
00:01:31,990 --> 00:01:34,190
scroll down and
16

17
00:01:34,360 --> 00:01:43,540
now what I'm gonna do is I'm going to copy this code here for our whale, come down here and paste it in.
17

18
00:01:44,920 --> 00:01:49,380
There's a few modifications that we can make here that will save us a bit of time.
18

19
00:01:49,390 --> 00:01:56,250
So first off we're gonna change the image to say thumbs up,
19

20
00:01:56,530 --> 00:02:03,810
then what we'll have to do is we'll have to generate the text as a string for the word cloud,
20

21
00:02:03,820 --> 00:02:04,110
right.
21

22
00:02:06,050 --> 00:02:13,700
In this case, I'm going to store all of this text in a variable called "ham_str" and that will
22

23
00:02:13,700 --> 00:02:24,030
be equal to "' '.join(flat_list_ham)".
23

24
00:02:24,770 --> 00:02:32,930
So we've got a list of non-spam words and we're going to join them all together into a single string. Before
24

25
00:02:32,930 --> 00:02:33,990
I run this,
25

26
00:02:34,040 --> 00:02:43,910
I have to do one last thing. I have to change this "novel_as_string" to our "ham_str" variable.
26

27
00:02:43,910 --> 00:02:47,120
That way I'm not generating another word cloud for Moby Dick,
27

28
00:02:47,160 --> 00:02:50,570
instead I'm generating it for non-spam emails.
28

29
00:02:52,810 --> 00:02:54,560
And here's our output.
29

30
00:02:54,670 --> 00:03:01,030
We still have the Moby Dick styling, but we can change that very easily with a different colour map and
30

31
00:03:01,030 --> 00:03:05,560
I think otherwise it looks pretty good. In terms of colour map,
31

32
00:03:05,590 --> 00:03:11,830
let's go for 'winter' and, I don't know, maybe a bit more granularity on the words.
32

33
00:03:14,350 --> 00:03:15,100
Now,
33

34
00:03:15,160 --> 00:03:22,660
I think this looks pretty good actually, but there's probably one thing that you'll notice that might
34

35
00:03:22,840 --> 00:03:24,520
bother you a little bit.
35

36
00:03:24,580 --> 00:03:28,330
"People", for example is spelled without the "e" at the end.
36

37
00:03:28,360 --> 00:03:34,830
"Change" is spelled without the "e" at the end. "Provide is spelled without the "e".
37

38
00:03:35,170 --> 00:03:37,700
We can only guess what this word "tri"
38

39
00:03:37,720 --> 00:03:43,990
is supposed to be. The reason for this is is that the data that we fed into our word cloud are the
39

40
00:03:43,990 --> 00:03:46,210
stemmed words.
40

41
00:03:46,210 --> 00:03:50,230
That's why we're seeing our word cloud presented like this.
41

42
00:03:50,710 --> 00:03:58,690
Our flattened list of non-spam keywords is coming from our nested list of non-spam keywords, which is
42

43
00:03:58,690 --> 00:04:06,760
coming from our nested list, which we've created up here when we applied our "clean_msg_
43

44
00:04:06,910 --> 00:04:09,400
no_html" method.
44

45
00:04:09,400 --> 00:04:12,610
This is where we did this stemming if you recall.
45

46
00:04:12,610 --> 00:04:17,920
If you wanted to see what this word cloud would look like without the stemmed words, then you'd have
46

47
00:04:17,920 --> 00:04:27,580
to copy this code, paste it in, use the original words and comment out
47

48
00:04:27,690 --> 00:04:38,250
this line right here. If you refresh this cell and then go to "Kernel" > "Restart & Run All", then you'll
48

49
00:04:38,250 --> 00:04:46,110
run all the cells in the entire notebook and after waiting a little while, you can scroll down and look
49

50
00:04:46,260 --> 00:04:53,210
at the output for the word cloud without the stemmed words.
50

51
00:04:53,220 --> 00:04:54,300
There we go.
51

52
00:04:54,300 --> 00:04:56,680
This is starting to look a lot better.
52

53
00:04:56,700 --> 00:05:03,150
The stemmed words are going to be verym very useful for our Bayes' classifier but they're actually not so
53

54
00:05:03,150 --> 00:05:05,480
useful for our word cloud.
54

55
00:05:05,580 --> 00:05:11,040
So if you've changed it like I did then it's important to make a mental note to change this back and
55

56
00:05:11,040 --> 00:05:17,680
use the stemmed words before we train our model. Now I've got one final challenge for you on the word
56

57
00:05:17,680 --> 00:05:19,510
cloud front.
57

58
00:05:19,510 --> 00:05:26,500
I'd like you to pull up the word cloud documentation and figure out how to use the custom font that
58

59
00:05:26,530 --> 00:05:33,910
I've included in the lesson resources instead of this default font here and then create a word cloud
59

60
00:05:34,330 --> 00:05:38,430
of all the words in the spam messages.
60

61
00:05:38,480 --> 00:05:40,440
I've included two font files.
61

62
00:05:40,630 --> 00:05:49,520
The "OpenSansCondensed-Bold" font file and I've included the "OpenSansCondensed-Light" font file.
62

63
00:05:49,840 --> 00:05:54,820
If you look at this word cloud here, what kind of font do you think would work best with it?
63

64
00:05:55,570 --> 00:06:02,650
What characteristics should a font have in order to make a word cloud look a lot more attractive? Because
64

65
00:06:02,800 --> 00:06:09,880
in my humble opinion this default font here doesn't work as well as it should.
65

66
00:06:09,880 --> 00:06:15,970
What kind of improvement on the font front would you make to this visualization?
66

67
00:06:15,970 --> 00:06:18,870
I can think of two improvements right off the bat.
67

68
00:06:19,030 --> 00:06:24,780
First off, the words should probably be written in all caps rather than lowercase.
68

69
00:06:25,450 --> 00:06:31,420
And second, the letters of the font should probably be closer together and bold.
69

70
00:06:31,540 --> 00:06:36,740
These are the first two things that I would try to make this look a lot more convincing.
70

71
00:06:37,210 --> 00:06:41,180
So I hope you will give this a shot when solving the challenge.
71

72
00:06:41,230 --> 00:06:46,930
Pause the video and have a go. All good?
72

73
00:06:46,930 --> 00:06:49,540
Here's the solution.
73

74
00:06:49,740 --> 00:06:57,220
We've already added the relative path to the thumbs down file so we can already use it.
74

75
00:06:57,460 --> 00:07:00,610
I'm going to copy this code right here
75

76
00:07:01,830 --> 00:07:05,070
and paste it in below.
76

77
00:07:05,070 --> 00:07:13,730
I'll swap up thumbs up for thumbs down, this line stays the same, as does this one. Our string of course
77

78
00:07:14,090 --> 00:07:22,280
will need to change, so this will be the spam string and this will come from the flattened spam list.
78

79
00:07:23,170 --> 00:07:29,660
And when it comes to generating the cloud we will use the spam string instead of the ham string. For a
79

80
00:07:29,660 --> 00:07:31,760
bit of visual differentiation,
80

81
00:07:31,760 --> 00:07:35,780
I'm going to go with something reddish from the color map.
81

82
00:07:35,780 --> 00:07:41,220
In fact, I'm going to try this "gist_heat" color map that we've got here.
82

83
00:07:41,330 --> 00:07:48,020
So instead of "winter" I'll go for "gist_heat" and I'll keep some of the other things the
83

84
00:07:48,020 --> 00:07:49,280
same.
84

85
00:07:49,370 --> 00:07:55,220
Now when I said check the word cloud documentation, what I wanted you to pick up on was the fact that
85

86
00:07:55,220 --> 00:08:06,180
you could supply a font here as an argument and you can do so by giving the word cloud a font path. The
86

87
00:08:06,180 --> 00:08:10,680
font path is the font that will be used in the word cloud.
87

88
00:08:10,860 --> 00:08:20,850
And this has to be a OTF or a TFF file and a TFF file is exactly what we've got. Going back up,
88

89
00:08:20,850 --> 00:08:29,730
I'll add my relative path here for my font file. So I'll call this one "CUSTOM_FONT_
89

90
00:08:29,730 --> 00:08:43,610
FILE" and this will be my "OpenSansCondensed-Bold.ttf" file.
90

91
00:08:43,920 --> 00:08:49,900
You can try both of these and see which one looks better, but I think I prefer the bold one. Just gotta
91

92
00:08:49,900 --> 00:08:57,540
make sure there's no typos here in this file name. Let me hit Shift+Enter on this, scroll back down,
92

93
00:08:58,720 --> 00:09:09,730
and now I can supply a "font_path" and set that equal to the custom font file.
93

94
00:09:09,800 --> 00:09:17,660
The other thing that you might have had to have a play with is the maximum font size. So in getting it to
94

95
00:09:17,660 --> 00:09:19,670
look the way you want it to,
95

96
00:09:19,670 --> 00:09:26,480
there is a mention of a font size parameter in the quick documentation as well. Despite the rather strange
96

97
00:09:26,480 --> 00:09:28,190
word wrapping at the moment,
97

98
00:09:28,190 --> 00:09:35,840
you can see it right here, "max_font_size" by default is equal to None, but we can give it another value
98

99
00:09:35,840 --> 00:09:46,160
as well, so "max_font_size" equals, say I don't know, I think 300 worked well last
99

100
00:09:46,160 --> 00:09:51,130
time I looked at this. Now let me hit Shift+Enter and see what this looks like.
100

101
00:09:52,460 --> 00:09:54,230
My computer is really struggling.
101

102
00:09:54,230 --> 00:09:56,800
I think I've got too many things open.
102

103
00:09:57,010 --> 00:09:57,380
Okay.
103

104
00:09:57,410 --> 00:09:58,610
So here we go.
104

105
00:09:58,640 --> 00:10:00,080
This is the result.
105

106
00:10:00,260 --> 00:10:07,520
I think the font looks pretty good, but I think we need to up the number of words being displayed in
106

107
00:10:07,520 --> 00:10:09,020
this image.
107

108
00:10:09,080 --> 00:10:14,110
Also we're still in lower case, so let's change these two things.
108

109
00:10:14,270 --> 00:10:21,350
I'm going to be pretty radical and change the maximum number of words to 2000 from 500 and when it
109

110
00:10:21,350 --> 00:10:29,560
comes to generating my word cloud I'm going to convert my string to uppercase letters.
110

111
00:10:29,810 --> 00:10:40,350
So "spam_str.upper()" will convert all of my letters to uppercase. Now I can refresh and
111

112
00:10:40,710 --> 00:10:43,410
see if we're happy with the result.
112

113
00:10:43,470 --> 00:10:49,240
I think the red color definitely worked really well but with these last two changes I think the design
113

114
00:10:49,240 --> 00:10:53,400
should really come together.
114

115
00:10:54,040 --> 00:10:55,960
Okay so that's interesting.
115

116
00:10:56,190 --> 00:11:00,480
The largest word has randomly been given a white color.
116

117
00:11:00,540 --> 00:11:06,350
This is probably not what I want, so I'm going to refresh this and see if I'm more lucky
117

118
00:11:06,390 --> 00:11:11,080
the second time round. Much better.
118

119
00:11:11,110 --> 00:11:11,740
Brilliant.
119

120
00:11:12,030 --> 00:11:18,800
So I quite like the look of this but I'd be quite curious what sort of designs you've come up with.
120

121
00:11:18,900 --> 00:11:25,920
And speaking of your own designs, a good question at this point is: Where can you find both fonts and
121

122
00:11:25,920 --> 00:11:31,770
masks for your own projects, for your own word clouds? When it comes to fonts,
122

123
00:11:31,780 --> 00:11:39,460
one of my favorite places to go to is Google Fonts, so Google Fonts are free and open source and you
123

124
00:11:39,460 --> 00:11:44,680
can just browse and download the font files that you like from here.
124

125
00:11:44,680 --> 00:11:49,670
This is a brilliant resource. When it comes to masks and icons,
125

126
00:11:49,780 --> 00:11:51,890
you've got a lot of choice as well.
126

127
00:11:51,910 --> 00:11:57,550
One place that you could check out is the icon set on fontawesome.com.
127

128
00:11:57,580 --> 00:12:03,910
Here you can download various icons that you can then use as masks for your PNG files.
128

129
00:12:03,940 --> 00:12:10,050
Just remember to double check the licensing on all of these image files that you're interested in using.
129

130
00:12:10,570 --> 00:12:15,430
In any case font awesome allows you to filter on free icons which is quite handy.
130

131
00:12:15,430 --> 00:12:22,660
So if we check out this accessible icon here, then you can download this icon as an SVG file, a vector
131

132
00:12:22,990 --> 00:12:28,510
which you can then scale as large as you like with some image processing software.
132

133
00:12:28,510 --> 00:12:32,800
Now with those tools in hand I think you're all set to create something fantastic.
133

134
00:12:32,830 --> 00:12:36,520
If you do please post it in the comments section below this video,
134

135
00:12:36,520 --> 00:12:40,420
I'd love to see what you've made. In the next lessons,
135

136
00:12:40,500 --> 00:12:47,800
we're going to generate the vocabulary and the dictionary for our Bayes' classifier. It's about time I get
136

137
00:12:47,800 --> 00:12:50,450
some food, so I'll see in the next lesson.
137

138
00:12:50,470 --> 00:12:51,510
Take care.