1
00:00:00,000 --> 00:00:03,040
Okay, let's get down to business.

2
00:00:03,570 --> 00:00:07,562
Using the simple diagram representing an ML

3
00:00:07,563 --> 00:00:11,796
system, we have input, output, and some

4
00:00:11,797 --> 00:00:14,690
brain with knowledge in the middle.

5
00:00:14,691 --> 00:00:19,844
In machine learning, features are the input data.

6
00:00:19,845 --> 00:00:25,290
A single feature represent a specific input variable.

7
00:00:25,291 --> 00:00:28,226
Think about the task of predicting

8
00:00:28,227 --> 00:00:31,530
the market price of a house.

9
00:00:31,531 --> 00:00:35,260
A single feature can be the size of

10
00:00:35,261 --> 00:00:38,448
that house, but this is not enough.

11
00:00:38,449 --> 00:00:41,568
We will need more relevant features that

12
00:00:41,569 --> 00:00:45,168
can better represent a house, so the

13
00:00:45,169 --> 00:00:48,350
system can better predict the market price.

14
00:00:48,351 --> 00:00:51,748
The features that can be used as input to

15
00:00:51,749 --> 00:00:55,332
such ML system can be the size, number of

16
00:00:55,333 --> 00:01:00,868
bedroom, street area, overall condition, and so on.

17
00:01:00,869 --> 00:01:05,069
Another example will be an email spam detector.

18
00:01:05,070 --> 00:01:08,760
And in that case, the features of such ML system

19
00:01:08,761 --> 00:01:12,558
can be the content of the email like the sender

20
00:01:12,559 --> 00:01:16,492
address, the time and the hour of the day

21
00:01:16,493 --> 00:01:21,868
the email was generated and sent, number of spammy words

22
00:01:21,869 --> 00:01:25,000
like free, money, and so on.

23
00:01:25,850 --> 00:01:30,422
In simple projects we can have 10-20 input

24
00:01:30,423 --> 00:01:33,856
features like we just saw, while in some

25
00:01:33,857 --> 00:01:38,822
very sophisticated machine learning project could use thousands

26
00:01:38,823 --> 00:01:41,818
and even millions of features.

27
00:01:41,819 --> 00:01:47,562
Mathematically, the list of input features will be represented

28
00:01:47,563 --> 00:01:50,788
as a vector with the size of n like

29
00:01:50,789 --> 00:01:54,370
x1, x2, x3, until xn.

30
00:01:55,830 --> 00:01:59,150
As the sentence is saying garbage

31
00:01:59,151 --> 00:02:02,230
in, we will get garbage out.

32
00:02:02,231 --> 00:02:04,776
We'll talk about it later in this training,

33
00:02:04,777 --> 00:02:08,892
but it is important to mention that selecting the

34
00:02:08,893 --> 00:02:13,800
right features is a critical step in the process.

35
00:02:14,330 --> 00:02:17,510
Something that is called feature selection.

36
00:02:18,510 --> 00:02:22,656
Okay, moving next, a label is the

37
00:02:22,657 --> 00:02:25,168
output of the machine learning system.

38
00:02:25,169 --> 00:02:27,072
It is the thing we would like

39
00:02:27,073 --> 00:02:30,900
to predict or classify using the system.

40
00:02:30,901 --> 00:02:34,900
After we train that system, a label could

41
00:02:34,901 --> 00:02:38,036
be the price of a real estate property.

42
00:02:38,037 --> 00:02:43,086
As we just talked about, the identified object in a picture

43
00:02:43,087 --> 00:02:46,664
like this is a dog, that's a cat, and so on.

44
00:02:46,665 --> 00:02:51,272
The type of incoming email, spam, not spam, the

45
00:02:51,273 --> 00:02:55,452
root cause of some events, the words being used

46
00:02:55,453 --> 00:02:58,330
in a video clip and much more.

47
00:02:58,331 --> 00:03:03,788
Mathematically, the label will be presented as Y as a

48
00:03:03,789 --> 00:03:07,960
simple math equation Y equal to the function of x.

49
00:03:09,710 --> 00:03:12,496
Now we can describe the meaning of

50
00:03:12,497 --> 00:03:15,750
examples, which is another key term.

51
00:03:15,751 --> 00:03:19,392
An example is a single instance of

52
00:03:19,393 --> 00:03:22,612
data represented by the letter x.

53
00:03:22,613 --> 00:03:25,114
There are two types of examples

54
00:03:25,115 --> 00:03:29,850
labeled examples and unlabeled examples.

55
00:03:29,851 --> 00:03:33,448
A labeled examples includes the features, which

56
00:03:33,449 --> 00:03:36,142
is the x and also the label

57
00:03:36,143 --> 00:03:39,750
related to such features which is Y.

58
00:03:39,751 --> 00:03:42,974
Think about an email, a user

59
00:03:42,975 --> 00:03:47,186
manually labeled as a spam.

60
00:03:47,187 --> 00:03:49,458
It is a labeled examples.

61
00:03:49,459 --> 00:03:52,946
Or think about an image as input,

62
00:03:52,947 --> 00:03:55,430
which is basically many small pixels.

63
00:03:56,010 --> 00:04:00,704
All those pixels are input features and

64
00:04:00,705 --> 00:04:03,616
inside the picture there is a specific

65
00:04:03,617 --> 00:04:07,550
object, like a specific type of animal.

66
00:04:07,551 --> 00:04:12,452
The label will be the name of that animal, okay,

67
00:04:12,453 --> 00:04:14,026
the type of that animal.

68
00:04:14,027 --> 00:04:17,476
If we have the image file and the type

69
00:04:17,477 --> 00:04:22,950
of animal inside, then we have one labeled example.

70
00:04:22,951 --> 00:04:26,152
If we have just the image file without the

71
00:04:26,153 --> 00:04:30,660
identified object inside, then we have an unlabeled example.

72
00:04:31,270 --> 00:04:35,532
A large group of labeled examples can be used

73
00:04:35,533 --> 00:04:39,240
as a training data set to train a model.

74
00:04:39,850 --> 00:04:44,348
If I will summarize we have features as input to

75
00:04:44,349 --> 00:04:50,034
the machine learning system, a label as the expected output.

76
00:04:50,035 --> 00:04:53,740
If we have a group of data instances that are

77
00:04:53,741 --> 00:04:59,452
already labeled, then they can be used as examples to

78
00:04:59,453 --> 00:05:03,890
train model, which is called a training data et.

79
00:05:03,891 --> 00:05:07,520
Moving next, let's talk about training a model.

80
00:05:07,521 --> 00:05:09,892
[No audio]