1
00:00:11,070 --> 00:00:14,180
Hey, everyone, and welcome back to TIME series analysis.

2
00:00:14,730 --> 00:00:20,640
So this lecture is the intro to the next section of the course, this section is going to be about how

3
00:00:20,640 --> 00:00:23,400
to apply machine learning models, two time series.

4
00:00:24,390 --> 00:00:26,320
Now, here's where the lines blur a little bit.

5
00:00:26,940 --> 00:00:32,100
Note that what we studied earlier, Arima, is technically no different from the machine learning models.

6
00:00:32,250 --> 00:00:33,360
We will study here.

7
00:00:33,900 --> 00:00:38,780
In fact, one of the models we will study is the AP model we looked at previously.

8
00:00:39,330 --> 00:00:44,670
So please keep this in mind that what is and what is not machine learning is just a subjective opinion.

9
00:00:45,120 --> 00:00:49,720
At the end of the day, if you understand how these things work, what you call it is irrelevant.

10
00:00:50,190 --> 00:00:55,230
Now, that being said, this section is more focused on models that would traditionally be taught in

11
00:00:55,230 --> 00:00:56,400
a machine learning course.

12
00:00:56,760 --> 00:01:02,700
Specifically, I've decided to focus on linear models, support vector machines and the random forest.

13
00:01:07,280 --> 00:01:12,440
So the main idea of this section is that you're not going to learn any machine learning model depth.

14
00:01:12,980 --> 00:01:17,810
This, of course, is about application two time series not deriving the duality theorem of support

15
00:01:17,810 --> 00:01:22,270
vector machines, nor understanding the statistics behind why the random force works.

16
00:01:22,760 --> 00:01:28,670
So to that end, the most important thing to focus on in this section is not any specific machine learning

17
00:01:28,670 --> 00:01:29,090
model.

18
00:01:30,110 --> 00:01:35,130
In fact, practically no discussion about any machine learning model is really necessary.

19
00:01:35,630 --> 00:01:40,940
What you should focus on in this section are the kinds of tasks that machine learning can solve and

20
00:01:40,940 --> 00:01:44,150
how we transform these tasks into time series tasks.

21
00:01:44,570 --> 00:01:47,330
So this process is completely model agnostic.

22
00:01:47,930 --> 00:01:53,210
That is, it doesn't really matter which machine learning models you use because the same rules apply

23
00:01:53,540 --> 00:01:56,690
and because it doesn't matter which machine learning models you use.

24
00:01:56,930 --> 00:02:01,530
Any discussion about a specific machine learning model is technically not needed.

25
00:02:02,390 --> 00:02:06,710
Now, that being said, I know that there will be at least a few students who would like to know a little

26
00:02:06,710 --> 00:02:08,950
something about the models will be using.

27
00:02:09,500 --> 00:02:12,750
So they will be discussed in a more intuition only setting.

28
00:02:13,910 --> 00:02:18,740
And again, if you want to learn these models in depth, you can always check the resources I've left

29
00:02:18,740 --> 00:02:20,490
in extra reading that text.

30
00:02:25,190 --> 00:02:27,510
So the outline for this section is pretty simple.

31
00:02:27,980 --> 00:02:32,690
We'll start with the two main categories of supervised learning, classification and regression.

32
00:02:33,560 --> 00:02:37,920
The important part of this is to understand these from a geometrical point of view.

33
00:02:38,390 --> 00:02:40,490
You'll learn one of my most important rules.

34
00:02:40,610 --> 00:02:43,070
Machine learning is nothing but geometry.

35
00:02:44,450 --> 00:02:49,310
Once you understand these concepts, well, then look at how to transform them into auto regressive

36
00:02:49,310 --> 00:02:50,030
models.

37
00:02:50,450 --> 00:02:54,380
These are somewhat like the AP models you studied earlier, but more powerful.

38
00:02:55,130 --> 00:03:00,470
The next step will be to quickly go over the intuition behind each machine learning model we will use.

39
00:03:01,220 --> 00:03:06,410
Again, this isn't necessary, but it should at least give you some intuition behind the libraries you'll

40
00:03:06,410 --> 00:03:07,100
be using.

41
00:03:07,940 --> 00:03:11,000
One important topic of this section is extrapolation.

42
00:03:11,660 --> 00:03:16,340
This is kind of another take on the concept of stationary, which we've looked at before.

43
00:03:17,090 --> 00:03:22,880
The reason this is important is people often have this unrealistic view that machine learning is magic.

44
00:03:23,190 --> 00:03:25,910
It can automatically learn any pattern from data.

45
00:03:26,390 --> 00:03:30,310
Well, what we will learn in this section is that there is a limit to this belief.

46
00:03:30,890 --> 00:03:35,620
This will further drive the point home that machine learning isn't magic, it's just geometry.

47
00:03:36,230 --> 00:03:41,270
After this important lesson, we'll be ready to apply what we've learned to several time series data

48
00:03:41,270 --> 00:03:41,870
sets.

49
00:03:42,730 --> 00:03:44,750
OK, so I hope you're excited for this section.

50
00:03:45,110 --> 00:03:47,330
Thanks for listening and I'll see you in the next lecture.