1
00:00:11,000 --> 00:00:16,760
So in this lecture, we will discuss a common misconception regarding how to forecast between the train

2
00:00:16,760 --> 00:00:17,570
and test sets.

3
00:00:18,260 --> 00:00:24,020
In these videos, we forecasted only four time steps within the test set for which we had true data.

4
00:00:24,740 --> 00:00:29,630
But what if we would like to forecast beyond this period for which we do not have true data?

5
00:00:30,350 --> 00:00:31,680
I'll present this as a quiz.

6
00:00:31,700 --> 00:00:36,980
So if you had this question, please think about it for a day or so and try to implement it yourself

7
00:00:37,250 --> 00:00:38,330
before moving on.

8
00:00:43,100 --> 00:00:43,460
Okay.

9
00:00:43,460 --> 00:00:45,530
So the answer is nothing changes.

10
00:00:46,070 --> 00:00:49,490
Remember that the model does not even use the test data for anything.

11
00:00:50,030 --> 00:00:54,990
So if you were thinking that the test data was used to make the forecast, this is incorrect.

12
00:00:55,550 --> 00:00:58,220
The test data is not used during the forecast.

13
00:00:58,610 --> 00:01:01,670
It is only used to compute our forecast metrics.

14
00:01:02,330 --> 00:01:07,970
In addition, remember that the whole purpose of splitting data into train and test or train and validation

15
00:01:08,300 --> 00:01:11,870
is to emulate how our model would perform on data out of sample.

16
00:01:12,440 --> 00:01:18,470
Therefore, the process should be exactly the same as how we would do a real forecast into the future,

17
00:01:18,830 --> 00:01:20,930
since that is what we are trying to emulate.