1
00:00:01,260 --> 00:00:04,740
Sometimes we have observations at wrong frequency.

2
00:00:06,200 --> 00:00:09,230
Maybe they are too fine or not fine.

3
00:00:09,280 --> 00:00:09,700
And of.

4
00:00:11,390 --> 00:00:19,760
For example, we may have sales data of every day and we may be trying to forecast next month's sales.

5
00:00:21,660 --> 00:00:27,230
Are we really having monthly sales values and we want to project next week sales?

6
00:00:29,050 --> 00:00:36,670
So sometimes the frequency of the data available does not match with the frequency at which we want

7
00:00:36,670 --> 00:00:37,480
to forecast.

8
00:00:39,680 --> 00:00:45,070
In such situations, we have to change the frequency of the data available with us.

9
00:00:46,700 --> 00:00:49,020
This is known as resampling de data.

10
00:00:52,090 --> 00:00:59,670
So in this video, you will discover how to increase and decrease the sampling frequency of the times

11
00:00:59,670 --> 00:01:00,240
of this data.

12
00:01:02,250 --> 00:01:09,660
After completing the section you below the two types of resampling and the main reason why you need

13
00:01:09,660 --> 00:01:10,670
to use them.

14
00:01:14,080 --> 00:01:20,360
As I told you, the resampling simply means changing the frequency of your time to these observations.

15
00:01:22,010 --> 00:01:27,290
Now, suppose you have quarterly data and you want to forecast monthly would fall.

16
00:01:29,090 --> 00:01:33,680
This means you will need to convert your quarterly data to monthly data.

17
00:01:35,720 --> 00:01:40,460
This means that you will have to add samples to the data to increase the frequency.

18
00:01:43,080 --> 00:01:49,230
Adding samples to data increases the frequency, and this is called Upsampling.

19
00:01:51,510 --> 00:01:53,460
This is one type of resampling.

20
00:01:57,090 --> 00:02:02,330
The other type is obviously downsampling in which we have more data.

21
00:02:03,430 --> 00:02:04,720
See quarterly data.

22
00:02:05,260 --> 00:02:07,320
And we want to make yearly predictions.

23
00:02:09,830 --> 00:02:16,190
In such a scenario, we have to reduce the number of samples and therefore reduce the frequency of our

24
00:02:16,190 --> 00:02:16,560
data.

25
00:02:18,300 --> 00:02:23,100
For example, you can see here we had quarterly data of three years.

26
00:02:24,990 --> 00:02:26,910
This is the footfall data for three years.

27
00:02:27,600 --> 00:02:35,480
We have to convert this quarterly data and do eyerly data so that we can predict the footfall in it

28
00:02:35,490 --> 00:02:35,790
for.

29
00:02:39,230 --> 00:02:42,560
In both cases, data must be invented.

30
00:02:43,980 --> 00:02:51,240
In case of Upsampling, we will calculate the fine grained observations using interpolation technique.

31
00:02:52,480 --> 00:02:59,830
And in case of downsampling, we will use some statistics to calculate the new aggregate values and

32
00:02:59,830 --> 00:03:00,680
the law.

33
00:03:00,940 --> 00:03:04,510
We have seen only one reason of doing these sampling.

34
00:03:05,200 --> 00:03:11,080
That is if the frequency of data is different from the frequency at which we want to make predictions.

35
00:03:12,540 --> 00:03:16,190
We will resemble our data to that frequency and then build our model.

36
00:03:18,140 --> 00:03:24,650
That is one more reason why we may do resampling, and that is for the purpose of feature engineering.

37
00:03:26,510 --> 00:03:33,350
For example, suppose we have monthly sales data and we want to predict future monthly sales.

38
00:03:35,020 --> 00:03:39,970
As I told you earlier in our forecasting model, we will input features.

39
00:03:41,690 --> 00:03:46,040
And one of the features could be Dodo teed off last year.

40
00:03:48,080 --> 00:03:52,100
Or it could be the total seed of the last day of the previous month.

41
00:03:54,140 --> 00:03:57,830
These features have different frequency level than the data.

42
00:04:00,420 --> 00:04:03,560
To create these features, we may do resampling.

43
00:04:06,570 --> 00:04:10,800
In this way, sometimes to get features we use resembling models until.

44
00:04:12,410 --> 00:04:15,290
Now let's see how to do resampling in our soffit.
