Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:11,100 --> 00:00:17,320
So in this lecture, we are going to discuss the types of tasks that you can do with Time series models.
2
00:00:17,970 --> 00:00:24,180
We'll start with the most intuitive task, which for most people is the one step forecast that is given.
3
00:00:24,180 --> 00:00:29,290
Why one and why two all the way up to what we would like to predict, YFC plus one.
4
00:00:29,310 --> 00:00:31,950
And so that's what we train our model to do.
5
00:00:33,150 --> 00:00:39,240
As a side note, be aware that in this class we might use X of T or Y of T to represent a time series.
6
00:00:39,600 --> 00:00:46,050
So in the context of a long time series, there won't be any difference between using X or Y in a supervised
7
00:00:46,050 --> 00:00:46,380
learning.
8
00:00:46,380 --> 00:00:47,470
This is a different story.
9
00:00:47,760 --> 00:00:49,860
So just keep this in the back of your mind.
10
00:00:54,340 --> 00:01:00,550
Now, the one step forecast is not the end of the story in practice, we often want to predict multiple
11
00:01:00,550 --> 00:01:01,790
steps into the future.
12
00:01:02,170 --> 00:01:04,130
We call this the forecast horizon.
13
00:01:04,870 --> 00:01:08,120
This is the length of time into the future that we would like to predict.
14
00:01:08,710 --> 00:01:12,990
For example, we might want to predict the sales of each day for the next week.
15
00:01:13,480 --> 00:01:16,010
One very realistic example is the weather.
16
00:01:16,450 --> 00:01:20,380
Imagine checking the Weather Channel and it only shows the weather one day ahead.
17
00:01:20,770 --> 00:01:23,770
This wouldn't be very useful compared to other weather channels.
18
00:01:24,310 --> 00:01:27,500
Now, this is not to say that one step forecasts are not useful.
19
00:01:27,730 --> 00:01:29,430
It all depends on context.
20
00:01:29,890 --> 00:01:34,990
For example, if you run a brick and mortar shop, you might want to forecast the sales of our products
21
00:01:35,050 --> 00:01:37,260
next month from monthly data.
22
00:01:37,810 --> 00:01:40,360
In this case, that is a one step forecast.
23
00:01:40,720 --> 00:01:45,800
This will help you purchase any inventory you might need to fulfill demand in the next month.
24
00:01:46,540 --> 00:01:48,850
And again, this all depends on context.
25
00:01:49,240 --> 00:01:53,010
When in doubt, ask your manager or your clients or your stakeholders.
26
00:01:57,670 --> 00:02:04,150
So in this course, we're going to learn two ways in which to produce multi-step forecast method number
27
00:02:04,150 --> 00:02:09,070
one is called the incremental forecast, which can be done with any one step predictor.
28
00:02:09,700 --> 00:02:15,250
Method number two is called the multi output forecast, which is limited to only certain models.
29
00:02:15,700 --> 00:02:21,700
For example, we'll study Arima, which can do one step forecasts but cannot do a multi output forecast.
30
00:02:22,150 --> 00:02:24,700
So let's discuss each of these models more in depth.
31
00:02:29,290 --> 00:02:31,890
For now, suppose that our model is a black box.
32
00:02:32,110 --> 00:02:34,360
It can only make a one step forecast.
33
00:02:34,870 --> 00:02:39,290
Let's suppose that in order to make this prediction, we use past data points.
34
00:02:39,670 --> 00:02:45,660
So given we of T minus P plus one up to Y of T, we can predict Y of T plus one.
35
00:02:46,330 --> 00:02:49,110
Let's call this prediction Y hat of T plus one.
36
00:02:50,020 --> 00:02:54,970
But suppose that our forecast horizon is three times steps so H is equal to three.
37
00:02:55,660 --> 00:03:01,020
In this case, what we can do is plug in our prediction as an input into our model.
38
00:03:01,600 --> 00:03:09,270
So for the next input will pass in Y of T minus plus two up to Y of T followed by Y had of T plus one.
39
00:03:09,940 --> 00:03:12,300
This will give us Y hat of T plus two.
40
00:03:13,030 --> 00:03:18,580
Now that we have our prediction for Y had of T plus two, we can take this, plug it into our input
41
00:03:18,580 --> 00:03:21,160
again and yet we have T plus three.
42
00:03:22,030 --> 00:03:27,430
The important thing to recognize is that we are not allowed to use the true values for YFC plus one
43
00:03:27,430 --> 00:03:33,970
and YFC plus two because the current time is only T, C plus one and C plus two are in the future.
44
00:03:38,680 --> 00:03:43,690
So just to give you a concrete example, suppose that we have we have one, we have two and we have
45
00:03:43,690 --> 00:03:46,240
three say that is equal to three.
46
00:03:46,240 --> 00:03:50,040
So our model uses three pass time points to predicts the next point.
47
00:03:50,650 --> 00:03:51,730
We'd like to predict why.
48
00:03:51,730 --> 00:03:52,270
Six.
49
00:03:52,390 --> 00:03:54,430
So our forecast horizon is three.
50
00:03:55,180 --> 00:03:57,820
Before we can do that, we must find we had four.
51
00:03:58,450 --> 00:04:00,160
So we use our model plugging in.
52
00:04:00,160 --> 00:04:01,810
Why one, why to and why three.
53
00:04:02,050 --> 00:04:03,640
And that gives us we had four.
54
00:04:04,210 --> 00:04:07,020
Now in order to estimate why five we plug in.
55
00:04:07,040 --> 00:04:07,630
Why two.
56
00:04:07,630 --> 00:04:08,170
Why three.
57
00:04:08,170 --> 00:04:09,400
And why had four.
58
00:04:09,820 --> 00:04:11,400
This gives us why had five.
59
00:04:12,070 --> 00:04:13,450
Finally we can plug in.
60
00:04:13,450 --> 00:04:14,170
Why three.
61
00:04:14,350 --> 00:04:16,540
We had four and we had five.
62
00:04:16,540 --> 00:04:18,380
And this will give us we had six.
63
00:04:19,390 --> 00:04:22,570
Now you might be wondering why can't you just plug in the true values.
64
00:04:22,690 --> 00:04:25,860
Why four and why five in order to estimate why six.
65
00:04:26,470 --> 00:04:32,260
The answer is if you imagine these are days to day is only day three in order to know the true value
66
00:04:32,260 --> 00:04:35,710
of why four and why five, we would have to wait until day five.
67
00:04:36,130 --> 00:04:39,220
And of course, in some businesses this might be unacceptable.
68
00:04:39,730 --> 00:04:45,160
As I always tell students who ask this question, imagine your boss asks you to make a forecast three
69
00:04:45,160 --> 00:04:45,850
days ahead.
70
00:04:46,300 --> 00:04:50,410
Your answer cannot be OK, but we just have to wait a two days to get an answer.
71
00:04:50,830 --> 00:04:52,720
As always, context is important.
72
00:04:52,930 --> 00:04:57,130
So ask your clients or your manager what your forecast horizon really is.
73
00:05:01,550 --> 00:05:07,090
Now, let's talk about method number two, method number two is the multi output forecast.
74
00:05:07,520 --> 00:05:11,180
Again, imagine that we have a black box model for some models.
75
00:05:11,180 --> 00:05:11,790
We will study.
76
00:05:11,820 --> 00:05:15,490
We'll see that they are capable of giving us multiple outputs at once.
77
00:05:16,040 --> 00:05:21,860
Therefore, if we have a forecast to rise Horizon H, then we can simply create a model with H outputs
78
00:05:21,860 --> 00:05:23,540
and obtain a forecast for each time.
79
00:05:23,540 --> 00:05:25,730
Step up to points in the future.
80
00:05:26,350 --> 00:05:30,650
As you can see, this is much simpler than building a forecast incrementally.
81
00:05:35,280 --> 00:05:41,700
The final task I want to discuss in this lecture is that of classification, this isn't normally discussed
82
00:05:41,700 --> 00:05:46,780
in a traditional time series analysis, but this being a more modern course, it's worth knowing.
83
00:05:47,430 --> 00:05:52,670
So in the previous examples, our job was always to predict a number in machine learning.
84
00:05:52,680 --> 00:05:57,210
We call this regression, but what if we'd like to predict a category instead?
85
00:05:58,050 --> 00:06:00,260
Again, let's pretend that we are neural link.
86
00:06:00,600 --> 00:06:05,940
We want to read a user's brain signals and yes, whether they are hungry or tired, hungry and tired,
87
00:06:05,940 --> 00:06:07,650
our categories, not numbers.
88
00:06:08,280 --> 00:06:13,980
Another example is taking motion readings from a smartphone and trying to guess what the user is doing
89
00:06:14,340 --> 00:06:16,250
walking, sleeping or sitting down.
90
00:06:16,740 --> 00:06:19,230
Again, these are categories, not numbers.
91
00:06:20,140 --> 00:06:25,630
OK, so these are some examples of the types of tasks that we can do with Time series analysis.
92
00:06:26,040 --> 00:06:28,200
Thanks for listening and I'll see you in the next lecture.
9093
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.