1
00:00:00,800 --> 00:00:04,250
In this lecture, we will learn how to implement moving evidence.

2
00:00:04,340 --> 00:00:06,920
Or am I model in Python?

3
00:00:10,170 --> 00:00:16,350
In Amman, Jordan, Verein auto regulation on the residual values of over time cities.

4
00:00:18,950 --> 00:00:22,110
So let's first create the recipe and values.

5
00:00:23,270 --> 00:00:27,860
We are going to create Tresillian values by implementing nine model first.

6
00:00:29,120 --> 00:00:30,640
So this is our dataset.

7
00:00:32,270 --> 00:00:33,630
We are using the same dataset.

8
00:00:36,470 --> 00:00:40,760
If you remember, we created my forecast, some lectors back.

9
00:00:40,970 --> 00:00:42,800
We are going to do the same thing here.

10
00:00:43,760 --> 00:00:50,180
We are creating a new column, which is speech by shifting over temperature, column by one record.

11
00:00:50,420 --> 00:00:52,830
So this is a leg, one record.

12
00:00:55,470 --> 00:00:58,340
Now, temp is the value we want to predict.

13
00:00:58,880 --> 00:01:00,800
And this is the X value.

14
00:01:02,510 --> 00:01:10,480
So if we just take or friends of them and pee, it will give us the serial values, flawed make model.

15
00:01:11,930 --> 00:01:18,230
If you are not comfortable with knife model or the computation that we are doing here, I recommend

16
00:01:18,230 --> 00:01:24,650
you to watch our name more than implementation in Python video first and then watch this video.

17
00:01:25,580 --> 00:01:28,700
So you have created the vestibules as well.

18
00:01:30,830 --> 00:01:34,610
So let's look at the first five and use of four data.

19
00:01:35,740 --> 00:01:36,880
First few of their day.

20
00:01:38,070 --> 00:01:39,480
These are the actual values.

21
00:01:40,120 --> 00:01:44,820
These are Democratic trade values, and their difference is that that's a and value.

22
00:01:46,790 --> 00:01:49,550
So we have implemented my forecast model.

23
00:01:50,120 --> 00:01:54,890
And we've also calculated great students from that model in our data.

24
00:01:59,500 --> 00:02:00,820
Known for moving average.

25
00:02:00,970 --> 00:02:03,280
We will run our total duration on this.

26
00:02:03,340 --> 00:02:04,510
That's the dual values.

27
00:02:07,060 --> 00:02:10,420
So let's first create tests and train data.

28
00:02:11,110 --> 00:02:14,560
We are selecting that third column from our D of data frame.

29
00:02:15,340 --> 00:02:23,320
And we are selecting last seven rows as our test data and the remaining rules as the train data since

30
00:02:23,320 --> 00:02:24,370
the first value is.

31
00:02:24,460 --> 00:02:33,420
And then we are ignoring it and we are saving all the records from the second one due to all the records

32
00:02:33,450 --> 00:02:36,170
except the last seven records as screen.

33
00:02:36,940 --> 00:02:41,860
And we are saving seven records as our test data.

34
00:02:45,000 --> 00:02:51,300
So let's trendies also, let's look at it first by Riluzole four green nitta.

35
00:03:02,370 --> 00:03:05,800
You can see we have all of these values here.

36
00:03:06,090 --> 00:03:10,310
And you can see that we have this sets of other values in our crane data.

37
00:03:12,730 --> 00:03:16,310
No, next, we have to implement auto regulation model on this.

38
00:03:16,450 --> 00:03:17,510
Rest of the world will lose.

39
00:03:18,130 --> 00:03:20,740
So we are importing it from its Smardon.

40
00:03:22,180 --> 00:03:30,070
We are creating a model object using function, and we are giving our train data as an input.

41
00:03:30,890 --> 00:03:33,010
And the next step, we are putting out more.

42
00:03:35,100 --> 00:03:43,570
Now, let's look at how many lagging variables are important in this auto regression on that Saleel

43
00:03:43,580 --> 00:03:43,910
model.

44
00:03:45,230 --> 00:03:50,240
You can see there won't be line lag where he will study important for this or politician.

45
00:03:50,990 --> 00:03:54,170
Now, we can also get the coefficient of all this variables.

46
00:03:54,890 --> 00:03:55,880
So you can see.

47
00:03:58,370 --> 00:03:59,570
These are the parameters.

48
00:03:59,900 --> 00:04:04,520
And remember, this Elvan value is the Legba value on Brazilian's.

49
00:04:05,890 --> 00:04:09,130
These are not the leg, the value of forward time series data.

50
00:04:09,220 --> 00:04:13,820
This out of the leg, the value of the silhouettes of night forecasts on a lot more.

51
00:04:14,800 --> 00:04:18,760
So we have coefficient of all the 29 leg variables.

52
00:04:20,350 --> 00:04:24,700
Now, if we just predict using a predict function of this model.

53
00:04:25,790 --> 00:04:30,830
We will be predicting the forecast, said glycerol for the next well do in time cities.

54
00:04:31,460 --> 00:04:38,120
But to get the forecast said value of all time cities, we have to add this forecast, said Tresillian

55
00:04:39,050 --> 00:04:39,830
in the forecast.

56
00:04:39,850 --> 00:04:43,940
Say night value that we have Vita's in this column.

57
00:04:46,440 --> 00:04:49,410
So let's spread the vestibule values.

58
00:04:49,920 --> 00:04:53,820
We are going to use the same model, underscore the fact, not predict.

59
00:04:55,790 --> 00:05:01,430
And we have to give a start, point and then point, we are giving the starting point as this Satoh

60
00:05:01,430 --> 00:05:08,990
forward test that is length of crane and the end point test length of crane plus length of test, minus

61
00:05:08,990 --> 00:05:09,200
one.

62
00:05:11,740 --> 00:05:12,380
And this.

63
00:05:14,850 --> 00:05:18,120
Now we have the forecast said, let's return value.

64
00:05:21,370 --> 00:05:22,930
The feed just right.

65
00:05:27,470 --> 00:05:28,000
First,

66
00:05:31,670 --> 00:05:37,280
you can see that we are getting the seven values this updo forecast said yesterday on the value.

67
00:05:39,410 --> 00:05:46,190
Now, we have to add this forecast said resolution values in the forecast that we have already made.

68
00:05:51,360 --> 00:05:53,630
What is the forecast that we have already made?

69
00:05:54,020 --> 00:05:54,740
That is.

70
00:05:55,960 --> 00:06:05,650
This so since we are using night forecast, which is stored in Colombe, we are getting the last seven

71
00:06:05,650 --> 00:06:06,690
records of day.

72
00:06:08,140 --> 00:06:14,390
You can see this at the forecasted values and these are the forecasted resolutions.

73
00:06:14,590 --> 00:06:22,430
So if we add this to data frame, we'll get the final forecasted value from our moving average model.

74
00:06:24,220 --> 00:06:25,870
So this is what we are doing here.

75
00:06:26,560 --> 00:06:31,780
We are creating another data frame that is predictions and we are adding these two values.

76
00:06:34,630 --> 00:06:39,490
Let's look at the values of this prediction.

77
00:06:40,150 --> 00:06:43,750
So these are the forecasted values from what, moving average more than.

78
00:06:46,290 --> 00:06:52,290
Let's calculate the MASC value, the MSU value is two.

79
00:06:53,580 --> 00:06:59,310
So if you remember from 079 forecast, we were getting masc value of around three point four.

80
00:07:00,030 --> 00:07:01,560
From what?

81
00:07:01,740 --> 00:07:07,540
Here model, we were getting masc value of one point five from over.

82
00:07:08,300 --> 00:07:11,030
We are more done with walk forward validation.

83
00:07:11,070 --> 00:07:18,390
We were getting masc value of around one point four and using this and more than we are getting an masc

84
00:07:18,390 --> 00:07:19,600
value of two.

85
00:07:21,600 --> 00:07:23,850
We can also lorden the same.

86
00:07:25,860 --> 00:07:33,390
Using the same port, you can see that this the blue line is the actual blue and red line is a credit

87
00:07:33,390 --> 00:07:33,960
card value.

88
00:07:37,220 --> 00:07:39,210
So this is all reimplement memorial.

89
00:07:39,970 --> 00:07:45,000
And by then I will repeat these steps once more.

90
00:07:45,620 --> 00:07:54,550
So first you have to use a model to create forecast set values, and then you'll need to create Tresillian

91
00:07:54,560 --> 00:07:56,720
values on those forecasted values.

92
00:07:57,230 --> 00:08:00,620
So the use may create the forecasted values.

93
00:08:01,250 --> 00:08:02,180
And then we.

94
00:08:04,010 --> 00:08:09,400
And then we took a defense off all night forecast and actual values to clear cultural values.

95
00:08:10,160 --> 00:08:17,300
We used it more than to train our model on the resolution values and to predict future forecasted reticule

96
00:08:17,300 --> 00:08:17,810
values.

97
00:08:18,770 --> 00:08:26,930
Then we again used our future forecast and values of our rules and added a two hour night forecast,

98
00:08:26,930 --> 00:08:30,420
said value to get the final forecast, said value trauma.

99
00:08:30,500 --> 00:08:31,490
What am I, modern?

100
00:08:36,020 --> 00:08:39,880
You can also implement walk forward on a memorial as well.

101
00:08:41,900 --> 00:08:48,160
You just have to run the same code on resolutions instead of actual values and then use those vestigial

102
00:08:48,280 --> 00:08:50,660
values to credit the final values.

103
00:08:53,010 --> 00:08:55,500
I recommend you to do that on your own.

104
00:08:56,190 --> 00:08:57,540
That's all for this video.

105
00:08:57,810 --> 00:08:58,180
Thank you.
