1
00:00:00,000 --> 00:00:02,640
Welcome to the fourth
and final course

2
00:00:02,640 --> 00:00:03,660
of this specialization,

3
00:00:03,660 --> 00:00:05,550
just one course
away from finishing

4
00:00:05,550 --> 00:00:07,230
this whole specialization and

5
00:00:07,230 --> 00:00:08,865
have you learnt a lot
about tensor flow.

6
00:00:08,865 --> 00:00:11,415
In this course, you learned
about sequence models,

7
00:00:11,415 --> 00:00:12,510
what are sequence models?

8
00:00:12,510 --> 00:00:14,160
So what we're going to be

9
00:00:14,160 --> 00:00:16,170
focusing on is one part
of sequence models,

10
00:00:16,170 --> 00:00:17,325
which is really time-series.

11
00:00:17,325 --> 00:00:19,350
Sequence models where
it's a case of if you can

12
00:00:19,350 --> 00:00:22,055
imagine a series of data
that changes over time.

13
00:00:22,055 --> 00:00:23,760
It might be the closing prices

14
00:00:23,760 --> 00:00:25,335
for stock on the stock exchange,

15
00:00:25,335 --> 00:00:26,745
or it could be weather.

16
00:00:26,745 --> 00:00:29,970
It could be how sunny it is
in California on a given day,

17
00:00:29,970 --> 00:00:31,290
or how rainy it is in Seattle

18
00:00:31,290 --> 00:00:32,640
on a given day,
that type of thing.

19
00:00:32,640 --> 00:00:34,410
So if you just imagine how

20
00:00:34,410 --> 00:00:36,210
an item of data changes

21
00:00:36,210 --> 00:00:38,260
over time and how it's
measured over time.

22
00:00:38,260 --> 00:00:40,560
So basically almost anything
like a spreadsheet,

23
00:00:40,560 --> 00:00:43,310
if I've a spreadsheet
where I have one day per

24
00:00:43,310 --> 00:00:47,865
row and two columns say one
for the California weather,

25
00:00:47,865 --> 00:00:50,475
one for the Seattle weather
to document how

26
00:00:50,475 --> 00:00:54,330
better my life is to yours in
Seattle than in California,

27
00:00:54,330 --> 00:00:59,565
and then we would have a new
network help us model that.

28
00:00:59,565 --> 00:01:01,760
Exactly. So we're
going to start by

29
00:01:01,760 --> 00:01:04,805
creating a synthetic
sequence of data,

30
00:01:04,805 --> 00:01:06,560
so that we can start
looking at what

31
00:01:06,560 --> 00:01:09,710
the common attributes that
you see in data series are.

32
00:01:09,710 --> 00:01:12,800
So for example, whether
data can be seasonal.

33
00:01:12,800 --> 00:01:15,410
It's sunnier in June
than it is in January

34
00:01:15,410 --> 00:01:18,190
or it's wetter in November
than it is in October,

35
00:01:18,190 --> 00:01:19,220
something along those lines.

36
00:01:19,220 --> 00:01:21,065
So you have that
seasonality of data.

37
00:01:21,065 --> 00:01:22,580
You can also, in some cases,

38
00:01:22,580 --> 00:01:24,020
have trends of data,

39
00:01:24,020 --> 00:01:26,630
like whether it probably doesn't
really trend although we

40
00:01:26,630 --> 00:01:27,860
could argue that it
strangely enough

41
00:01:27,860 --> 00:01:29,360
idea with climate change,

42
00:01:29,360 --> 00:01:31,190
but like a stock data
may trend upwards

43
00:01:31,190 --> 00:01:33,320
over time or downwards
over some other times,

44
00:01:33,320 --> 00:01:35,330
and then of course
the random factor

45
00:01:35,330 --> 00:01:37,280
that makes it hard
to predict is noise.

46
00:01:37,280 --> 00:01:39,330
So you can have
like seasonal data,

47
00:01:39,330 --> 00:01:40,460
you can have trends in your data,

48
00:01:40,460 --> 00:01:42,560
but then you can have noise
on that data as well.

49
00:01:42,560 --> 00:01:45,800
So that the average
temperature of a Tuesday

50
00:01:45,800 --> 00:01:49,325
in June and California
might be 85 degrees,

51
00:01:49,325 --> 00:01:51,350
but it might be 85.5 degrees,

52
00:01:51,350 --> 00:01:52,985
it might be 84.5 degrees.

53
00:01:52,985 --> 00:01:54,890
So you get that noise
in the data.

54
00:01:54,890 --> 00:01:56,390
So we want to start

55
00:01:56,390 --> 00:01:58,370
looking at various methods
that can be used

56
00:01:58,370 --> 00:02:00,590
statistically in
the Machine Learning to help us

57
00:02:00,590 --> 00:02:03,760
predict data given
seasonality trend and noise.

58
00:02:03,760 --> 00:02:06,420
Then in this course, at
the end of this course,

59
00:02:06,420 --> 00:02:08,465
one of the most cool applications

60
00:02:08,465 --> 00:02:11,035
is to use these ideas
to model sunspot.

61
00:02:11,035 --> 00:02:14,600
Sunspot activity. Yeah.
So sunspot activity

62
00:02:14,600 --> 00:02:15,725
is really interesting

63
00:02:15,725 --> 00:02:18,020
because the sun has

64
00:02:18,020 --> 00:02:19,550
an 11-year cycle although

65
00:02:19,550 --> 00:02:21,565
some astronomers tell me
it's a 22-year cycle,

66
00:02:21,565 --> 00:02:23,060
there's actually
two 11-year cycles

67
00:02:23,060 --> 00:02:24,715
like nestled beside each other.

68
00:02:24,715 --> 00:02:26,735
Will we resolve this
in this course?

69
00:02:26,735 --> 00:02:28,000
That remains to be seen.

70
00:02:28,000 --> 00:02:29,210
You're going to have to study

71
00:02:29,210 --> 00:02:31,110
all the way through
and then we'll see,

72
00:02:31,110 --> 00:02:32,780
but the idea then is that you

73
00:02:32,780 --> 00:02:34,405
do get that nice seasonality,

74
00:02:34,405 --> 00:02:37,010
and we have data measuring

75
00:02:37,010 --> 00:02:39,845
back about 250 years worth
of sunspot activity.

76
00:02:39,845 --> 00:02:42,050
So that's on a monthly basis
counting the number

77
00:02:42,050 --> 00:02:44,225
of sunspots that had been
spotted by astronomers.

78
00:02:44,225 --> 00:02:45,740
So we do definitely see

79
00:02:45,740 --> 00:02:48,470
that 11 year cycle or
maybe the 22-year cycle,

80
00:02:48,470 --> 00:02:50,270
and there's a lot
of noise in there,

81
00:02:50,270 --> 00:02:51,800
the seasonality and that stuff.

82
00:02:51,800 --> 00:02:52,970
So it's fun to build

83
00:02:52,970 --> 00:02:54,605
something to protect
sunspot activity.

84
00:02:54,605 --> 00:02:58,490
In fact, sunspot activity is
very important to NASA and

85
00:02:58,490 --> 00:03:00,080
other space agencies because

86
00:03:00,080 --> 00:03:02,900
it affects satellite operations.

87
00:03:02,900 --> 00:03:04,490
So in this course,

88
00:03:04,490 --> 00:03:07,415
you'll start by learning
about sequence models,

89
00:03:07,415 --> 00:03:08,780
a time series data,

90
00:03:08,780 --> 00:03:10,640
first practicing
these skills and building

91
00:03:10,640 --> 00:03:12,680
these models on artificial Data,

92
00:03:12,680 --> 00:03:14,090
and then at the end
of this course,

93
00:03:14,090 --> 00:03:16,580
you get to take all these ideas
and apply them to

94
00:03:16,580 --> 00:03:19,460
the exciting problem of
molding sunspot activity.

95
00:03:19,460 --> 00:03:20,935
So let's get started.

96
00:03:20,935 --> 00:03:23,170
Please go on to the next video.