1
00:00:00,660 --> 00:00:06,030
In this video, we will learn how to implement walk forward validation for a demo model.

2
00:00:08,100 --> 00:00:10,410
So let's first aim for it.

3
00:00:10,620 --> 00:00:12,520
We are going to use the same dataset.

4
00:00:13,440 --> 00:00:14,860
That is simple dots.

5
00:00:14,900 --> 00:00:15,260
Yes.

6
00:00:16,440 --> 00:00:18,000
Let's look at the data

7
00:00:20,880 --> 00:00:22,000
here.

8
00:00:22,290 --> 00:00:25,040
You can see that we have the month value and the sales value.

9
00:00:26,250 --> 00:00:29,640
So this month, values are four 19 zero one.

10
00:00:30,630 --> 00:00:41,580
So if you see one nine zero is missing from this data and that twice, pandas will not be able to identify

11
00:00:41,580 --> 00:00:41,850
that.

12
00:00:41,880 --> 00:00:42,990
This is the date for me.

13
00:00:43,860 --> 00:00:47,570
So for this example, you can also use the time it.

14
00:00:48,120 --> 00:00:52,570
So first you have to define a password using this segment.

15
00:00:53,580 --> 00:00:54,300
In this case.

16
00:00:54,360 --> 00:01:01,620
And then you can also give this password as an argument while you are loading CSP in that way.

17
00:01:01,710 --> 00:01:03,360
You will have a date.

18
00:01:03,360 --> 00:01:07,630
Time for my trade data in your first column instead of this month value.

19
00:01:08,700 --> 00:01:12,240
But for now, be our only focus on the sales value.

20
00:01:12,360 --> 00:01:15,090
And we are running our model only on the sales value.

21
00:01:15,150 --> 00:01:17,560
That's why we are not going to change our.

22
00:01:21,560 --> 00:01:24,830
So let the wait over there trying to test and train.

23
00:01:26,360 --> 00:01:33,560
So we will take 70 percent of our data and do training and remaining 30 percent of the time to test.

24
00:01:34,280 --> 00:01:39,720
So first, I'm defining or train size and then I'm dividing my dataset.

25
00:01:39,830 --> 00:01:42,170
That is the year to test and train.

26
00:01:45,770 --> 00:01:55,850
Let's look at the shape of data so you can see that we have a long 25 records and green and their own

27
00:01:56,190 --> 00:02:00,340
290 records and our best for this.

28
00:02:00,860 --> 00:02:01,380
So, yeah.

29
00:02:01,580 --> 00:02:05,310
Bill, I don't believe in the courts and test and 25, the courts and train.

30
00:02:05,530 --> 00:02:07,340
So we have 36 records.

31
00:02:09,710 --> 00:02:16,550
No implementing walk forward for a remise almost similar to the more forward validation we implemented

32
00:02:16,550 --> 00:02:22,550
for year, more than we are willing to use the same, except first we are defining data.

33
00:02:23,570 --> 00:02:27,050
Then we will run the loop for all the values in test.

34
00:02:29,000 --> 00:02:31,610
We will create a model for each.

35
00:02:31,610 --> 00:02:39,020
Then in this fall loop will save the first value all forward via data into a list called Predict.

36
00:02:40,400 --> 00:02:48,560
And at the end of each one, we will include the test value and to dream data that we are going to use

37
00:02:48,590 --> 00:02:51,040
to dream of what model in the next one.

38
00:02:53,630 --> 00:02:56,180
So first we are initializing our data.

39
00:02:56,180 --> 00:02:56,600
We will?

40
00:02:56,600 --> 00:03:01,610
S train variable and we are also initializing or predict list.

41
00:03:02,990 --> 00:03:04,820
Then 40 in test.

42
00:03:06,020 --> 00:03:09,560
So this for loop will go on for eleven values of test.

43
00:03:10,520 --> 00:03:13,300
First we are defining or more than that is Atima.

44
00:03:13,610 --> 00:03:17,690
We are training this model on data and we are using BBC.

45
00:03:17,690 --> 00:03:21,010
You and you find common one, common Xeloda respectively.

46
00:03:22,160 --> 00:03:24,290
Then in the next step we are fitting or model.

47
00:03:25,470 --> 00:03:29,310
Then we are forecasting the next value using weight.

48
00:03:29,340 --> 00:03:32,060
We do model under Scott Fed dort forecast.

49
00:03:32,780 --> 00:03:42,320
And since Y is a combination of three arrays, we are just extracting the value of forecasts from this

50
00:03:42,320 --> 00:03:42,740
way.

51
00:03:43,650 --> 00:03:45,080
I concede almost zero.

52
00:03:46,550 --> 00:03:48,780
We are upending this VI forecast.

53
00:03:49,340 --> 00:03:55,460
And to predict function and then we are also spending our data in the next cycle.

54
00:03:55,520 --> 00:03:57,460
New data will going to bring my model.

55
00:03:58,130 --> 00:04:00,970
And it will predict the next value of fly.

56
00:04:02,810 --> 00:04:05,690
So let's run this.

57
00:04:08,490 --> 00:04:16,890
You can see that we have eleven values, so this is the prediction four times better P plus one is the

58
00:04:16,950 --> 00:04:20,940
prediction for time 30 plus two is the prediction for time.

59
00:04:20,940 --> 00:04:28,020
Predict a plus three using or third model is the prediction for time period D plus for using our fourth

60
00:04:28,020 --> 00:04:29,290
model and so on.

61
00:04:30,180 --> 00:04:36,690
So we have implemented walk forward evaluation on our data using my model.

62
00:04:37,290 --> 00:04:41,010
Now let's look at this list as well.

63
00:04:42,810 --> 00:04:44,490
This will contain the same data.

64
00:04:46,720 --> 00:04:51,170
Now to look at the MSCI value we will import.

65
00:04:51,970 --> 00:04:54,330
I mean, squared error from Eskinder.

66
00:04:55,350 --> 00:05:01,890
And then we will give the actual values or test that is test taught values and the predicted values

67
00:05:01,950 --> 00:05:04,050
that we have in our product list.

68
00:05:05,230 --> 00:05:08,240
So if you run this, this is the MASC value.

69
00:05:08,870 --> 00:05:12,300
You can compare this MASC value with other models as well.

70
00:05:14,650 --> 00:05:19,500
And so we implement Wolke forward validation on a premium order.

71
00:05:21,810 --> 00:05:23,370
That's all for this lecture.

72
00:05:23,490 --> 00:05:23,940
Thank you.