1
00:00:10,520 --> 00:00:14,480
So in this lecture, we're going to discuss a somewhat subtle topic.

2
00:00:14,720 --> 00:00:20,960
So students who are new to machine learning can often conflate the concepts of prediction and modeling.

3
00:00:21,200 --> 00:00:26,150
Most beginners to machine learning understand the need to make predictions, but not so much they need

4
00:00:26,150 --> 00:00:26,940
to model.

5
00:00:26,960 --> 00:00:28,760
So what's the difference?

6
00:00:28,970 --> 00:00:31,400
Well, prediction is exactly how it sounds.

7
00:00:31,400 --> 00:00:36,080
It's about making predictions or in other words, trying to guess something which you do not know for

8
00:00:36,080 --> 00:00:36,740
sure.

9
00:00:36,860 --> 00:00:38,660
This makes intuitive sense.

10
00:00:38,660 --> 00:00:43,580
So, for example, we might want to know whether or not someone will click our ad, We want to know

11
00:00:43,580 --> 00:00:48,050
how high or low a stock price will go, or we want to find objects in an image.

12
00:00:48,050 --> 00:00:50,840
These are all examples of making predictions.

13
00:00:54,970 --> 00:00:57,190
But so what does it mean to model?

14
00:00:57,430 --> 00:01:00,740
Modeling gives a functional form to your time series.

15
00:01:00,760 --> 00:01:02,500
This is very important.

16
00:01:02,590 --> 00:01:08,230
Not only does it tell you how to make predictions, but it also gives you deeper insight into why the

17
00:01:08,230 --> 00:01:10,210
Time series is the way it is.

18
00:01:10,300 --> 00:01:13,180
It helps you answer questions about your time series.

19
00:01:13,210 --> 00:01:19,360
For example, you can explain why a Time series is mean reverting, or you can explain why time series

20
00:01:19,360 --> 00:01:20,860
will grow unbounded.

21
00:01:21,250 --> 00:01:25,840
Very importantly, it can also tell you how predictable a Time series is.

22
00:01:26,230 --> 00:01:31,210
Someone who doesn't know how to model might waste their time trying to predict something which is actually

23
00:01:31,210 --> 00:01:32,170
unpredictable.

24
00:01:32,560 --> 00:01:37,960
For example, we all know intuitively that trying to predict a coin flip would not be useful.

25
00:01:38,140 --> 00:01:43,540
Someone who understands how to model will realize that what they have is unpredictable and won't waste

26
00:01:43,540 --> 00:01:45,610
time trying to make better predictions.

27
00:01:50,000 --> 00:01:54,380
In the Field of Time series we might call making Predictions Forecasting.

28
00:01:54,560 --> 00:01:58,760
So time series forecasting is the act of making predictions.

29
00:01:58,970 --> 00:02:02,480
However, modeling would be time series analysis.

30
00:02:02,600 --> 00:02:07,520
In any case, this distinction is not too important for this course, but pay attention to which is

31
00:02:07,520 --> 00:02:09,470
which as we go through the course.

32
00:02:09,620 --> 00:02:12,590
Understand what you are doing and why you are doing it.

33
00:02:12,620 --> 00:02:18,710
Sometimes the goal is to make an accurate forecast, but sometimes the goal is to gain a better understanding

34
00:02:18,710 --> 00:02:20,480
of the data we're working with.

35
00:02:20,480 --> 00:02:23,210
And modeling is one way to help us do this.