0
1
00:00:00,240 --> 00:00:10,000
So the next step is to use CreateML to train our machine learning model. And in order to do that,
1

2
00:00:10,140 --> 00:00:17,480
we, of course, have to open up our brand-new Xcode 10. And as you can see, at the moment, I'm working with
2

3
00:00:17,490 --> 00:00:24,000
Version 10 beta 3, and this is because I'm recording this ahead of time so that you'll be able to get
3

4
00:00:24,000 --> 00:00:30,300
started using these brand-new tools and brand-new APIs and frameworks as soon as possible.
4

5
00:00:30,300 --> 00:00:37,650
Now, if you're following along with the tutorials before September which is when the grandmaster or the
5

6
00:00:37,650 --> 00:00:40,650
GM version of Xcode 10 gets released,
6

7
00:00:40,830 --> 00:00:48,600
so Apple normally releases the official version sometime around September, late September. And you'll
7

8
00:00:48,600 --> 00:00:53,950
know that it's the official version because you'll be able to download it from the App Store.
8

9
00:00:53,980 --> 00:00:59,640
Now, if you're trying to follow along with the tutorials before you're able to download Xcode 10 off
9

10
00:00:59,640 --> 00:01:06,450
the App Store, then you would have noticed a note in the last lesson where I described how you can get
10

11
00:01:06,450 --> 00:01:14,400
hold of the beta version of Xcode 10 so that you can follow along with me even before the official version
11

12
00:01:14,400 --> 00:01:15,120
gets released
12

13
00:01:15,120 --> 00:01:21,330
out to the public. So as a rule of thumb, if you're doing this tutorial in October, then you should be
13

14
00:01:21,330 --> 00:01:26,010
going to the App Store and downloading the latest version of Xcode 10.
14

15
00:01:26,250 --> 00:01:31,830
If you're doing this before October, then you need to follow the instructions that I've given in order
15

16
00:01:31,830 --> 00:01:35,240
to download the beta version of Xcode 10.
16

17
00:01:35,340 --> 00:01:41,470
So once you have done though, then you should be able to open it up, and you can run Xcode beta alongside
17

18
00:01:41,550 --> 00:01:43,050
with the normal Xcode.
18

19
00:01:43,050 --> 00:01:44,470
So that's not a problem.
19

20
00:01:44,490 --> 00:01:47,190
You don't have to give up one for the other.
20

21
00:01:47,190 --> 00:01:55,710
So, now we're on our familiar welcome screen and we're going to create a new playground. Now whereas usually
21

22
00:01:55,740 --> 00:01:59,030
we've been creating playgrounds for iOS.
22

23
00:01:59,250 --> 00:02:06,000
Now in this case, in order to be able to use CreateML to build our machine learning models,
23

24
00:02:06,120 --> 00:02:11,070
we have to do that using the power of our laptop.
24

25
00:02:11,070 --> 00:02:20,010
So we need to click on the macOS tab to create a brand-new Blank playground for macOS.
25

26
00:02:20,340 --> 00:02:27,130
And so we're going to go ahead and click Next and you can save it anywhere you wish and give it a name.
26

27
00:02:27,420 --> 00:02:35,940
So I'm gonna call my playground ImageRecognitionModelMaker.
27

28
00:02:35,940 --> 00:02:42,240
So once you've created this brand-new macOS playground, the first thing that we're going to do is to
28

29
00:02:42,240 --> 00:02:47,610
make sure that we've enabled manual running of our playground.
29

30
00:02:47,610 --> 00:02:55,350
So you might have seen this in previous lessons. But down here, there is a little play button. And when
30

31
00:02:55,350 --> 00:02:59,820
you press on it, it will execute all the code inside your playground.
31

32
00:02:59,820 --> 00:03:07,850
Now, if you click on it and hold, then you will get to a menu with two options.
32

33
00:03:07,890 --> 00:03:15,210
Now, you can either automatically run your playground so it runs all the code every few seconds or whenever
33

34
00:03:15,210 --> 00:03:19,520
there's changes, or you can select manually run.
34

35
00:03:19,710 --> 00:03:26,070
And this will allow you to specify the parts of your code that you want to run and when you want to
35

36
00:03:26,070 --> 00:03:26,910
run it.
36

37
00:03:26,910 --> 00:03:34,710
Now, when we're creating our machine learning model, we're going to train our data and save it to a location.
37

38
00:03:35,010 --> 00:03:41,160
We don't want that to happen every single time we write a new line of code or every few seconds when
38

39
00:03:41,160 --> 00:03:43,910
it decides to automatically run our code.
39

40
00:03:43,950 --> 00:03:48,620
So we're going to make sure that the Manually \Run part is selected.
40

41
00:03:48,690 --> 00:03:55,320
And when you do that, you'll see this little play button show up inside the left margin.
41

42
00:03:55,320 --> 00:03:59,540
And that means that we can click on it to run, for example, only line 1.
42

43
00:03:59,640 --> 00:04:06,000
If I click on it here or if I click on it here, it'll run lines 1 through to four, and that gives us much
43

44
00:04:06,000 --> 00:04:09,510
more fine-grained control over what happens in our playground.
44

45
00:04:10,080 --> 00:04:15,930
So, now that we've done that, we're just gonna go ahead and delete all of the existing code that's inside our
45

46
00:04:15,930 --> 00:04:24,490
playground, and I'm going to import not just CreateML, but I'm going to import CreateMLUI.
46

47
00:04:24,720 --> 00:04:33,360
And this is a framework that allow us to create a user interface inside playgrounds so that we can vastly
47

48
00:04:33,360 --> 00:04:41,400
simplify the process of creating a image recognition machine learning model down to just writing two
48

49
00:04:41,400 --> 00:04:42,670
lines of code.
49

50
00:04:42,750 --> 00:04:45,260
And let me show you what those two lines are.
50

51
00:04:45,270 --> 00:04:51,660
So the first thing we need to do is to create a new image classifier builder object.
51

52
00:04:51,780 --> 00:05:01,000
So I'm going to use a constant and I'm gonna call it builder and it's going to be set to equal MLImageClassifierBuilder.
52

53
00:05:01,000 --> 00:05:01,600
.
53

54
00:05:02,410 --> 00:05:08,980
And I'm going to initialize it as a new builder object.
54

55
00:05:08,980 --> 00:05:17,410
Now, once I've got that object, then I can simply say builder dot, and I'm gonna call the method called
55

56
00:05:17,560 --> 00:05:21,600
showInLiveView. And that's it.
56

57
00:05:21,640 --> 00:05:30,550
That's all the code that we need to build a image classification user interface that allow us to build
57

58
00:05:31,120 --> 00:05:32,790
our image recognition model.
58

59
00:05:32,860 --> 00:05:41,140
So the next step, once we've written all of that code, is to switch on the assistant editor or the split
59

60
00:05:41,140 --> 00:05:42,020
screen.
60

61
00:05:42,220 --> 00:05:48,160
And this, on the right side, is our live view, as you can see here.
61

62
00:05:48,610 --> 00:05:51,410
And when we click on this button,
62

63
00:05:51,430 --> 00:05:58,720
so the run button, somewhere below the line where we've got our builder.showInLiveView,
63

64
00:05:59,020 --> 00:06:02,080
so we want to run all three lines of code, basically.
64

65
00:06:02,230 --> 00:06:09,100
And if I click it here, then it'll run all the code from line 1 through to 7, so that it can import
65

66
00:06:09,110 --> 00:06:10,540
CreateMLUI.
66

67
00:06:10,540 --> 00:06:18,550
It will create a new instance of MLImageClassifierBuilder and it will run the method to show the
67

68
00:06:18,550 --> 00:06:20,920
builder in the live view.
68

69
00:06:20,950 --> 00:06:23,050
So let's go ahead and give that a spin.
69

70
00:06:29,930 --> 00:06:39,410
Now, depending on the age of your laptop and the amount of rRAM that you have in your laptop, then this
70

71
00:06:39,410 --> 00:06:43,750
can take anywhere between just under a minute to maybe five minutes.
71

72
00:06:43,850 --> 00:06:47,240
But be patient and just let it do its thing.
72

73
00:06:47,240 --> 00:06:50,770
Once it's ready, you'll see that it says, "Ready to continue
73

74
00:06:50,780 --> 00:06:52,360
image recognition model maker."
74

75
00:06:52,820 --> 00:06:58,820
And inside the live view on the right side here, you can see we have this dotted line to allow us to
75

76
00:06:58,820 --> 00:07:02,410
start dropping images to begin training.
76

77
00:07:02,420 --> 00:07:10,850
So this is a good cue for us to go ahead and find our Training Data folder and easiest ways just to
77

78
00:07:10,850 --> 00:07:12,590
have it here on your desktop.
78

79
00:07:13,130 --> 00:07:20,210
And we're going to drag the entire folder of Training Data which, remember, currently, has three subfolders
79

80
00:07:20,530 --> 00:07:22,610
called Cat, Dog, and Rabbit,
80

81
00:07:22,610 --> 00:07:27,920
each containing a number of images that represent those labels.
81

82
00:07:28,010 --> 00:07:33,770
So we're going to take this entire folder called Training Data and we're just going to drop it into
82

83
00:07:33,830 --> 00:07:36,660
this dotted line area.
83

84
00:07:37,040 --> 00:07:43,730
And now you can see something pretty crazy and pretty special goes on in the background, as the machine learning
84

85
00:07:43,760 --> 00:07:52,670
image classifier builder goes through looking at all of these images trying to identify the features
85

86
00:07:52,760 --> 00:07:55,790
that are unique to each folder.
86

87
00:07:55,790 --> 00:08:03,500
What is unique about a rabbit or what is unique about the image of a cat and try to see how it can classify
87

88
00:08:03,800 --> 00:08:06,590
and learn what each of these animals look like.
88

89
00:08:06,590 --> 00:08:14,850
Now, once that is done, you will get a percentage up here and you'll see a success optimal solution found,
89

90
00:08:15,020 --> 00:08:21,800
that means that everything was successful, and we can go about going to the next part. And you can see
90

91
00:08:21,800 --> 00:08:27,030
down here, it says, "Drop images in here to begin testing."
91

92
00:08:27,050 --> 00:08:34,940
So, now this is our cue to grab our Testing Data set which, remember, again, contains folders that have
92

93
00:08:34,970 --> 00:08:41,770
our cat, dog, and rabbit each with around four or five images, around 20 percent of our Training Data
93

94
00:08:41,770 --> 00:08:50,540
set, and we're just going to drag and drop it into this area here. And now our model is going to go through
94

95
00:08:50,600 --> 00:08:53,480
those testing images,
95

96
00:08:53,540 --> 00:09:00,920
use what it's learned from our Training Data to try and classify those images. And you can see if you
96

97
00:09:01,100 --> 00:09:06,910
scroll down over here in the Live View, you can see that the prediction was there should be a cat
97

98
00:09:07,070 --> 00:09:14,220
and it checks the label or the answer, and it says, "I got it right," "I got it right," I got it right."
98

99
00:09:14,240 --> 00:09:18,400
And you can look through it's testing of itself.
99

100
00:09:18,470 --> 00:09:22,400
This is kind of like when you have flashcards and you test yourself on whether if you've got the right
100

101
00:09:22,430 --> 00:09:23,270
answer.
101

102
00:09:23,270 --> 00:09:28,790
That's basically what it's doing here. And in our case, we've got a 100 percent evaluation.
102

103
00:09:29,270 --> 00:09:31,170
So this number is important.
103

104
00:09:31,430 --> 00:09:38,030
Whenever you train your image classifier or any sort of machine learning model, you want to see what
104

105
00:09:38,030 --> 00:09:45,560
was the success rate of when we tested our model against images that it's never seen before.
105

106
00:09:45,560 --> 00:09:51,710
Now, of course, if you show it the same image that you had in the Training Data set, it already knows the
106

107
00:09:51,710 --> 00:09:52,250
answer,
107

108
00:09:52,280 --> 00:09:59,600
so that won't be a test. But if you show it images that it's never seen before, but of the same object
108

109
00:09:59,930 --> 00:10:06,710
and it classifies them correctly, then that gives you a higher evaluation score.
109

110
00:10:06,710 --> 00:10:13,520
Now, if you find that based off the data that you're trying to train your machine learning model on, that
110

111
00:10:13,520 --> 00:10:17,920
your evaluation score or your testing score is quite low,
111

112
00:10:17,960 --> 00:10:22,400
so, say, below 80 percent, or if it varies a lot,
112

113
00:10:22,430 --> 00:10:24,740
so sometimes it's 20 percent,
113

114
00:10:24,830 --> 00:10:31,220
other times, it's 80 percent, then that means that you don't have enough data.
114

115
00:10:31,220 --> 00:10:38,720
So for each of your categories or each of your labels, you need to provide more data until you can reach
115

116
00:10:38,780 --> 00:10:40,370
a higher accuracy.
116

117
00:10:40,400 --> 00:10:48,290
So usually that involves taking the object, say, a cat, and finding images of different types of cats, or
117

118
00:10:48,290 --> 00:10:57,290
cats taken from different angles or with different rotations, and basically providing it with more data,
118

119
00:10:57,620 --> 00:11:04,100
so that it can learn the intricate differences so that you can learn what are the intricate features
119

120
00:11:04,460 --> 00:11:07,540
that makes a cat a cat.
120

121
00:11:07,550 --> 00:11:14,600
So, for example, if you were to train your machine learning model, not just on cards, but on different species
121

122
00:11:14,600 --> 00:11:21,140
of cats, so they can recognize the species, then you'll need to provide a lot more data in order to get
122

123
00:11:21,140 --> 00:11:23,690
it to a satisfactory evaluation score.
123

124
00:11:23,690 --> 00:11:26,180
So something between 80 and 100.
124

125
00:11:26,300 --> 00:11:33,910
So within that short period of time of writing very little code and dragging and dropping some images,
125

126
00:11:34,000 --> 00:11:40,450
we've trained an image classifier that now can recognize a cat, a dog, and a rabbit.
126

127
00:11:40,510 --> 00:11:46,800
So all we have to do now is we can rename our ImageClassifier by double-clicking up here.
127

128
00:11:47,050 --> 00:11:49,350
So I'm going to give it a name that makes sense.
128

129
00:11:49,390 --> 00:11:59,960
So I'm going to call it my PetImageClassifier and you can edit the author or the description, or the
129

130
00:11:59,960 --> 00:12:03,950
license, the version, and also specify where you want to save it.
130

131
00:12:04,070 --> 00:12:07,640
So I'm just gonna save it inside my desktop.
131

132
00:12:09,970 --> 00:12:12,220
And I'm going to click Save.
132

133
00:12:12,220 --> 00:12:21,280
So, now if we head over to our desktop, then you can see we have our brand-new .mlmodel file that
133

134
00:12:21,280 --> 00:12:28,720
is our pet image classifier. And that's all that we have to do in order to do something that used to
134

135
00:12:28,720 --> 00:12:34,540
be so complex that we need to rely on IBM to create it for us,
135

136
00:12:34,540 --> 00:12:40,410
and we've just done it using our own laptop and only two lines of code.
136

137
00:12:40,480 --> 00:12:43,180
So that is so incredibly cool.
137

138
00:12:43,300 --> 00:12:51,580
And in the next lesson, we're going to be using this pet image classifier to build our own image recognition
138

139
00:12:51,610 --> 00:12:55,310
app using our own data and our own model.
139

140
00:12:55,330 --> 00:12:58,450
So for all of that and more, I'll see you on the next lesson.