1
00:00:00,210 --> 00:00:02,670
Okay, so now let's talk about Amazon SageMaker.

2
00:00:02,670 --> 00:00:05,480
So SageMaker is a fully managed service

3
00:00:05,480 --> 00:00:07,280
for developers and data scientists

4
00:00:07,280 --> 00:00:08,950
to build machine learning model.

5
00:00:08,950 --> 00:00:10,760
So all of the services so far we've seen

6
00:00:10,760 --> 00:00:12,130
in this section that are all

7
00:00:12,130 --> 00:00:14,180
managed machine learning service with

8
00:00:14,180 --> 00:00:15,770
a very specific purpose,

9
00:00:15,770 --> 00:00:18,140
for example translates some text,

10
00:00:18,140 --> 00:00:19,500
transcribes some audio,

11
00:00:19,500 --> 00:00:21,130
or convert text into audio

12
00:00:21,130 --> 00:00:22,450
or analyze parts of a text,

13
00:00:22,450 --> 00:00:25,220
but SageMaker is a higher level

14
00:00:25,220 --> 00:00:26,980
machine learning service where you have

15
00:00:26,980 --> 00:00:29,070
your actual developers or your data scientists

16
00:00:29,070 --> 00:00:31,720
within your organization create

17
00:00:31,720 --> 00:00:33,260
and build machine learning model.

18
00:00:33,260 --> 00:00:34,900
So it is a lot more involved

19
00:00:34,900 --> 00:00:37,180
and a lot more difficult to use.

20
00:00:37,180 --> 00:00:40,090
Now, when you want to do this kind of processes

21
00:00:40,090 --> 00:00:41,920
to build a machine learning model,

22
00:00:41,920 --> 00:00:43,640
you have to do a bunch of steps that I will show

23
00:00:43,640 --> 00:00:44,630
you in a second,

24
00:00:44,630 --> 00:00:46,200
and all these are quite difficult to do

25
00:00:46,200 --> 00:00:47,080
in one place,

26
00:00:47,080 --> 00:00:49,500
plus you need to provision some servers

27
00:00:49,500 --> 00:00:52,420
to perform these competitions to create these models,

28
00:00:52,420 --> 00:00:54,380
and that can be cumbersome as well.

29
00:00:54,380 --> 00:00:56,150
So this is where SageMaker comes in.

30
00:00:56,150 --> 00:00:58,860
So SageMaker will try to help you all along the way

31
00:00:58,860 --> 00:01:00,010
for the process.

32
00:01:00,010 --> 00:01:02,595
So I will show you what machine learning actually is,

33
00:01:02,595 --> 00:01:04,209
and this is a simplified version,

34
00:01:04,209 --> 00:01:05,500
if you are a machine learning expert

35
00:01:05,500 --> 00:01:07,500
don't get mad at me please,

36
00:01:07,500 --> 00:01:09,380
but let's say you wanted to build a model,

37
00:01:09,380 --> 00:01:11,170
or let's say I wanted to build a model

38
00:01:11,170 --> 00:01:13,380
to predict what score you're going to get

39
00:01:13,380 --> 00:01:15,780
at your certified CLAP practitioner exam.

40
00:01:15,780 --> 00:01:17,020
So how will I do it?

41
00:01:17,020 --> 00:01:18,900
Say for example that I am a developer

42
00:01:18,900 --> 00:01:19,970
or a data scientist,

43
00:01:19,970 --> 00:01:22,528
so I'm going to gather all your data from

44
00:01:22,528 --> 00:01:24,840
the actual performance of my students.

45
00:01:24,840 --> 00:01:26,707
So I will ask maybe 10,000 students

46
00:01:26,707 --> 00:01:29,680
to give me information about how many years of experience

47
00:01:29,680 --> 00:01:30,750
in IT they had,

48
00:01:30,750 --> 00:01:32,680
how many years of experience with AWS they had,

49
00:01:32,680 --> 00:01:34,100
how much time they spent on the course,

50
00:01:34,100 --> 00:01:36,500
how many practice exams they did, etc. etc,

51
00:01:36,500 --> 00:01:38,880
so I'll try to gather as much data as possible,

52
00:01:38,880 --> 00:01:40,340
and then I'm going to label the data.

53
00:01:40,340 --> 00:01:42,780
So that means you need to say which columns

54
00:01:42,780 --> 00:01:43,630
corresponds to what,

55
00:01:43,630 --> 00:01:45,753
and also you need to give some kind of score,

56
00:01:45,753 --> 00:01:48,200
and the score is the actual score for me

57
00:01:48,200 --> 00:01:49,680
at the exam that someone obtains.

58
00:01:49,680 --> 00:01:50,910
For example someone did not pass,

59
00:01:50,910 --> 00:01:54,160
670, maybe he didn't do the course completely,

60
00:01:54,160 --> 00:01:55,320
that would be the reason.

61
00:01:55,320 --> 00:01:57,610
Maybe someone passed with high grade,

62
00:01:57,610 --> 00:02:02,610
so 990 or maybe someone with an even higher grade 934,

63
00:02:03,000 --> 00:02:06,464
so, each student will get a specific score,

64
00:02:06,464 --> 00:02:10,080
and my guess is that based on the data that I've collected

65
00:02:10,080 --> 00:02:12,250
I can predict what the score will be.

66
00:02:12,250 --> 00:02:13,610
So I've first done the labeling,

67
00:02:13,610 --> 00:02:15,760
and that labeling process can be quite complicated to do

68
00:02:15,760 --> 00:02:16,960
in practice.

69
00:02:16,960 --> 00:02:18,820
Then I need to build a machine learning model,

70
00:02:18,820 --> 00:02:21,530
which is how can I predict these scores

71
00:02:21,530 --> 00:02:23,570
from the historical data.

72
00:02:23,570 --> 00:02:25,040
So, then you build that machine learning model

73
00:02:25,040 --> 00:02:26,480
and then you have to train it

74
00:02:26,480 --> 00:02:27,313
and tune it.

75
00:02:27,313 --> 00:02:29,610
So this is another part that's actually quite difficult

76
00:02:29,610 --> 00:02:32,270
to do, which is how to refine my model over time

77
00:02:32,270 --> 00:02:36,250
to better fit my data and my outputs.

78
00:02:36,250 --> 00:02:39,210
Okay, so now SageMaker and all of this will help

79
00:02:39,210 --> 00:02:40,100
you with the labeling,

80
00:02:40,100 --> 00:02:41,440
the building, the training

81
00:02:41,440 --> 00:02:43,030
and tuning, but not only.

82
00:02:43,030 --> 00:02:44,760
Now we have a machine learning model

83
00:02:44,760 --> 00:02:45,800
and it is created,

84
00:02:45,800 --> 00:02:46,850
it's fully working.

85
00:02:46,850 --> 00:02:48,350
So now I need to use it.

86
00:02:48,350 --> 00:02:50,210
So this is called deploying machine learning models.

87
00:02:50,210 --> 00:02:52,670
So, we're going to get new data coming in.

88
00:02:52,670 --> 00:02:54,500
For example you are the new student,

89
00:02:54,500 --> 00:02:55,780
and I'm going to survey you,

90
00:02:55,780 --> 00:02:56,613
and I'm going to say okay,

91
00:02:56,613 --> 00:02:58,100
how many years of experience do you have in IT,

92
00:02:58,100 --> 00:03:00,400
with AWS, how much time have you spent on this course,

93
00:03:00,400 --> 00:03:02,830
and then I will apply based on this data

94
00:03:02,830 --> 00:03:03,810
you just given me,

95
00:03:03,810 --> 00:03:06,240
I will apply the machine learning model that I have created

96
00:03:06,240 --> 00:03:07,220
from before.

97
00:03:07,220 --> 00:03:08,940
And then the machine learning model will say,

98
00:03:08,940 --> 00:03:12,350
for example, hey it sends based on the data you have

99
00:03:12,350 --> 00:03:14,250
I'm going to predict that this student

100
00:03:14,250 --> 00:03:17,350
will pass with a score of 906.

101
00:03:17,350 --> 00:03:18,910
And this whole process,

102
00:03:18,910 --> 00:03:20,240
of labeling, building the model,

103
00:03:20,240 --> 00:03:21,310
training it, tuning it,

104
00:03:21,310 --> 00:03:24,530
applying it can be all done within SageMaker.

105
00:03:24,530 --> 00:03:26,570
So, that's it for a quick introduction,

106
00:03:26,570 --> 00:03:27,420
but I hope you liked it,

107
00:03:27,420 --> 00:03:29,370
and I will see you in the next lecture.