1
00:00:11,110 --> 00:00:16,660
So one of the reasons I kept this example relatively simple is that I really want you to test your own

2
00:00:16,660 --> 00:00:18,010
ideas out for yourself.

3
00:00:18,640 --> 00:00:23,680
You might be wondering what if I added in different features like prices of other stocks or foreign

4
00:00:23,680 --> 00:00:29,230
exchange or even the weather, or maybe adding different features is useless and it's better to only

5
00:00:29,230 --> 00:00:30,770
have the Target Time series.

6
00:00:31,330 --> 00:00:34,780
What if you would use auto or hyper parameter optimization?

7
00:00:35,020 --> 00:00:36,610
Will that improve your results?

8
00:00:37,180 --> 00:00:40,820
These are all excellent questions to answer using what you've learned.

9
00:00:41,440 --> 00:00:44,900
Basically, the number of experiments you can do is combinatorial.

10
00:00:45,550 --> 00:00:51,910
This section was focused on showing you how to use the US forecast service so that you know how it works

11
00:00:52,120 --> 00:00:54,630
using almost the simplest example possible.

12
00:00:55,240 --> 00:01:00,340
The simplest example would be a one time series with no additional features, but I thought it would

13
00:01:00,340 --> 00:01:03,100
be useful to include the related Time series as well.

14
00:01:03,310 --> 00:01:04,960
So you can see how it works.

15
00:01:09,590 --> 00:01:16,250
So the exercise for this section is this first and foremost, if you have any questions about what if

16
00:01:16,250 --> 00:01:21,560
I tried this or what if I tried that, then the answer is, as always, to actually do it.

17
00:01:22,460 --> 00:01:27,370
The only way you will know how something will perform is if you actually try it and look at the output.

18
00:01:28,100 --> 00:01:32,900
If you have a favorite time series that you would like to forecast and you think an industrial strength

19
00:01:32,900 --> 00:01:37,270
enterprise solution like a US forecast will help give that a try.

20
00:01:41,980 --> 00:01:47,860
I also encourage you to extend the example we did in various ways, as mentioned, the possibilities

21
00:01:47,860 --> 00:01:51,060
are really combinatorial, but here are some ideas.

22
00:01:51,820 --> 00:01:52,150
No.

23
00:01:52,150 --> 00:01:54,790
One, try multiple stock tickers at once.

24
00:01:55,300 --> 00:01:58,270
As you recall, we only use the S&P 500.

25
00:01:58,660 --> 00:02:03,790
It would be interesting to see whether performance improves when we learn from several stock price time

26
00:02:03,790 --> 00:02:05,470
series instead of just one.

27
00:02:06,970 --> 00:02:11,920
No, to try the same example with stock returns instead of stock prices.

28
00:02:12,280 --> 00:02:19,930
Well, stationery help number three, try hourly data instead of daily data, see if more granularity

29
00:02:20,110 --> 00:02:21,480
helps improve accuracy.

30
00:02:22,720 --> 00:02:25,210
Number four, try exogenous data.

31
00:02:25,570 --> 00:02:32,380
For example, use another API to grab news headlines or Twitter feeds and maybe perform sentiment analysis

32
00:02:32,380 --> 00:02:32,980
as well.

33
00:02:34,150 --> 00:02:39,940
OK, so I hope this gave you some ideas for what you might want to try in order to complete this exercise.

34
00:02:40,450 --> 00:02:43,860
Really, the goal is to use this for something which you are interested in.

35
00:02:44,110 --> 00:02:47,780
So if you have any burning questions or ideas you think might work.

36
00:02:48,040 --> 00:02:49,770
Now is the time to test them out.