1
00:00:00,360 --> 00:00:07,290
In this lecture, we are going to learn about how to create a remote model in Python and how to select

2
00:00:07,350 --> 00:00:10,920
B, B and Q values for A more modern.

3
00:00:13,510 --> 00:00:18,190
So for this problem, we are going to use shampoo, not CSB dataset.

4
00:00:19,360 --> 00:00:23,080
You can download this dataset from the resources section of this video.

5
00:00:23,650 --> 00:00:29,650
So let's import this and let's take the first five videos of ordinator say.

6
00:00:32,760 --> 00:00:40,080
So we have month column and we have sales column and we want to create our model on the sales column

7
00:00:40,080 --> 00:00:40,830
of this data.

8
00:00:42,380 --> 00:00:43,210
Now, let's.

9
00:00:45,610 --> 00:00:47,160
Glad to see sales data.

10
00:00:51,060 --> 00:00:53,100
You can see this is the sales data.

11
00:00:54,000 --> 00:00:56,970
We have data offered on prettify records.

12
00:01:00,500 --> 00:01:10,160
Now, here, if you look at it, you can see that there is no seasonality at such and there is some

13
00:01:10,160 --> 00:01:11,730
sort of quarterdeck crane.

14
00:01:12,350 --> 00:01:14,790
At first the increase is very slow.

15
00:01:15,500 --> 00:01:20,870
But as we go on, the the increase is increasing with the time.

16
00:01:21,650 --> 00:01:26,210
So this is some sort of order decreeing that we are seeing in the sales data.

17
00:01:28,070 --> 00:01:32,840
And therefore, since this is according to Crane, we need to defensing.

18
00:01:34,160 --> 00:01:35,610
To make it stationary.

19
00:01:37,440 --> 00:01:48,210
That's why our B should be Q so in Atima, we have three parameters, B, B and Q B sense for a number

20
00:01:48,210 --> 00:01:48,820
of items.

21
00:01:48,960 --> 00:01:56,940
We want to consider for authorization, B sense for a number of defensing and Q sense for a number of

22
00:01:56,940 --> 00:02:00,060
items we want to consider for moving average part.

23
00:02:03,600 --> 00:02:11,760
Since our data has quadratic Krang, we will be using Bique what to do in case of linear trend.

24
00:02:11,910 --> 00:02:13,470
We should use the equal to one.

25
00:02:15,750 --> 00:02:20,280
Now let's look at how to select B and Q parameters.

26
00:02:23,670 --> 00:02:27,500
So to select you, you should Blondeau Autocorrelation Klodt.

27
00:02:29,940 --> 00:02:34,120
We have already discussed this graph and times it is with Malaysian Park.

28
00:02:34,890 --> 00:02:37,890
So we are just going to take.

29
00:02:43,990 --> 00:02:47,310
This is the auto relation part for the sales data that we have.

30
00:02:49,250 --> 00:02:51,800
We have the correlation well-used with the lag very well.

31
00:02:51,890 --> 00:02:55,230
So this is the correlation value with leg one.

32
00:02:55,280 --> 00:02:58,460
This is the correlation value with leg two and so on.

33
00:02:59,660 --> 00:03:02,630
We have this data for around today, five leg values.

34
00:03:06,210 --> 00:03:09,900
We have confidence interval mount as a strict length.

35
00:03:12,450 --> 00:03:16,700
In this graph, you should look at the value of weight.

36
00:03:17,170 --> 00:03:19,750
The line is crossing the confidence interval.

37
00:03:20,680 --> 00:03:26,410
So if you can see the line is crossing confidence interval line around here.

38
00:03:27,760 --> 00:03:30,820
So the value just before this point is five.

39
00:03:31,480 --> 00:03:35,620
So we should consider five as the value for you.

40
00:03:37,660 --> 00:03:45,310
So for any time, cities, if you want to run a remote more than first, block the line, check and

41
00:03:45,310 --> 00:03:46,510
identify the train.

42
00:03:49,030 --> 00:03:51,420
Then not autocorrelation lord.

43
00:03:53,380 --> 00:04:00,240
And look at where the autocorrelation plot is crossing the confidence interval line.

44
00:04:02,990 --> 00:04:07,190
And then also flawed, the partial autocorrelation draft to find out.

45
00:04:07,260 --> 00:04:11,680
Be here, the line is crossing after black file.

46
00:04:11,850 --> 00:04:14,690
That's why we should select you to fight.

47
00:04:15,950 --> 00:04:18,890
No, let's applaud autocorrelation Lord.

48
00:04:19,550 --> 00:04:22,730
Autocorrelation Lord is available in a sex model.

49
00:04:23,690 --> 00:04:30,930
So first let's imported and then Lord take again.

50
00:04:31,010 --> 00:04:34,120
These are the autocorrelation values with the leg variable.

51
00:04:35,630 --> 00:04:38,780
We have considered for thin legs for this graph.

52
00:04:39,350 --> 00:04:41,900
And we're loading the parcel.

53
00:04:42,040 --> 00:04:49,160
Although coordination dropped on the sales data, the blue band here is again the confidence interval.

54
00:04:50,810 --> 00:04:55,790
Now you can see that this partial autocorrelation is crossing the confidence interval.

55
00:04:55,910 --> 00:04:57,900
Add value three two.

56
00:04:57,980 --> 00:05:03,760
Therefore, we should consider two as be for our remodel.

57
00:05:06,320 --> 00:05:09,110
The first barrier is it's for the very well itself.

58
00:05:09,170 --> 00:05:13,100
So that's why we are getting partial out of religion value of one.

59
00:05:13,250 --> 00:05:14,460
This is for legman.

60
00:05:14,510 --> 00:05:15,580
This is for like two.

61
00:05:15,620 --> 00:05:16,840
And this is phylactery.

62
00:05:18,770 --> 00:05:22,130
The graph is crossing confidence interval between two and three.

63
00:05:22,280 --> 00:05:23,800
That's why we should consider two.

64
00:05:27,940 --> 00:05:36,800
Now, since we have the value of B, Q and B, we should train over at a memorial on these parameters.

65
00:05:37,420 --> 00:05:40,770
So first we are importing an e-mail from sex model.

66
00:05:41,750 --> 00:05:43,530
So it's imported.

67
00:05:44,710 --> 00:05:46,990
Then on the other steps are almost similar.

68
00:05:47,110 --> 00:05:48,970
We first define our model object.

69
00:05:49,570 --> 00:05:52,420
Our model is a rhema on sales data.

70
00:05:53,110 --> 00:05:57,520
And we have to give believe to values using this parameter.

71
00:05:57,730 --> 00:06:00,280
So we are using order equal to.

72
00:06:00,430 --> 00:06:02,860
And then we can pass a double first.

73
00:06:02,890 --> 00:06:05,500
We have to give B, then B, then Q.

74
00:06:06,700 --> 00:06:09,970
That's why we are passing a couple of come to my life.

75
00:06:11,470 --> 00:06:12,550
Let's run this.

76
00:06:12,990 --> 00:06:14,720
Now let's fit our model than.

77
00:06:20,740 --> 00:06:24,580
So the effort and the model ignored ignore this morning, saw this.

78
00:06:24,690 --> 00:06:25,330
Messages.

79
00:06:27,550 --> 00:06:31,970
Now to look at all local fishing and their values.

80
00:06:32,440 --> 00:06:35,430
You can use modern fact dot somebody's mental.

81
00:06:36,160 --> 00:06:42,040
So if you run this, you will get all the information that you need.

82
00:06:42,730 --> 00:06:45,550
So here we have the modern name.

83
00:06:45,700 --> 00:06:47,350
The number of observations.

84
00:06:48,400 --> 00:06:51,430
And then here you can see the consent value.

85
00:06:52,150 --> 00:06:53,950
The variables that we are considering.

86
00:06:54,370 --> 00:06:59,080
So we are considering five leg variables for auto regression.

87
00:06:59,710 --> 00:07:09,190
That's why we have five variables named as one and two and three and four and five with prefix as air.

88
00:07:09,390 --> 00:07:10,990
Other stand for auto regulation.

89
00:07:12,850 --> 00:07:18,370
Then our cue is to that is we are considering two variables for moving average.

90
00:07:19,360 --> 00:07:25,960
That's why we have two wearables, L1, L2 with refix Emma Emma sense for moving average.

91
00:07:27,160 --> 00:07:31,540
You can also see the coefficient values for all these variables.

92
00:07:31,600 --> 00:07:35,890
You can also see the P values for these variables.

93
00:07:36,190 --> 00:07:41,410
The smaller the Bulu, the more important that variable is.

94
00:07:41,830 --> 00:07:51,760
So the most important variable in our model is L1 and a little four auto regulation and L2 for moving

95
00:07:51,760 --> 00:07:52,180
average.

96
00:07:55,270 --> 00:08:00,160
So using the P value, you can also define importance of variables.

97
00:08:05,500 --> 00:08:12,470
Now, we will also look at the resolution where news of more than you can get that has real value from

98
00:08:12,470 --> 00:08:13,150
door dressing.

99
00:08:13,250 --> 00:08:13,630
My take.

100
00:08:14,030 --> 00:08:18,970
So I'm just equating Rassa realistic way to model fake thought.

101
00:08:19,560 --> 00:08:22,910
President elect Flodden, this desolate.

102
00:08:26,070 --> 00:08:31,590
Now, we have discussed that the red civil should be in the form of white noise.

103
00:08:32,040 --> 00:08:38,150
There should be no visible pattern in your resolution if there is any pattern in your resolution.

104
00:08:38,190 --> 00:08:42,810
That means that you can improve your model and there is some information in your model that you are

105
00:08:42,810 --> 00:08:43,290
missing.

106
00:08:45,450 --> 00:08:50,800
So you can see that there is no trend or seasonality that is going on in our lives.

107
00:08:50,920 --> 00:08:53,940
Well, and this is acting like a white noise.

108
00:08:54,690 --> 00:08:57,570
We can also look at the mean value of the restaurant.

109
00:08:59,130 --> 00:09:07,860
You can see that the mean value of resolution is eleven and the minimum, 25, 50 percent and 75 percentile.

110
00:09:08,340 --> 00:09:10,870
And Maxwell, you are following the bell curve.

111
00:09:11,730 --> 00:09:16,950
So the sort of a 10 restaurant statistics are looking good.

112
00:09:16,980 --> 00:09:21,210
And you should also look at the restaurant value after running at him more than.

113
00:09:25,180 --> 00:09:27,970
Now, these are some of the video agents offering my more than.

114
00:09:30,400 --> 00:09:35,900
We just feels this one, Atima, where we're defined to be and be.

115
00:09:40,660 --> 00:09:48,200
Now, if you want to run auto regression, you can use a rhema where the variables for Amet is zero.

116
00:09:48,730 --> 00:09:55,840
And if you want to just use moving average, you can define number of auto technician terms to be use

117
00:09:55,840 --> 00:09:57,190
and model as zero.

118
00:09:59,410 --> 00:10:08,290
So if you just give P and the values and Q is zero, then you are running auto regression model.

119
00:10:09,520 --> 00:10:17,650
And if you're PS zero and you are giving one lady and Q A loose, then you are running.

120
00:10:17,890 --> 00:10:18,520
And my model.

121
00:10:21,980 --> 00:10:29,620
So there is no need of finding out the rest of the year and running auto regression on that to execute

122
00:10:29,630 --> 00:10:33,550
moving average more than you can just use a remote model as well.

123
00:10:33,820 --> 00:10:35,400
Do random moving average model.

124
00:10:37,700 --> 00:10:39,110
So an animal model.

125
00:10:39,260 --> 00:10:40,670
All three models are covered.

126
00:10:40,700 --> 00:10:45,970
You can always use a remote more than with BD and you will use.

127
00:10:46,370 --> 00:10:53,500
You can use auto regression with you equate to zero and you can also use moving average with be quick

128
00:10:53,540 --> 00:10:53,990
Lozito.

129
00:10:58,120 --> 00:11:02,300
Now, he might have to mentor spoof forecast of future Trelew.

130
00:11:02,730 --> 00:11:04,620
First is the forecast method.

131
00:11:05,490 --> 00:11:11,280
If you just rate model for DORT forecasts, it will automatically give you the forecast for next Piter.

132
00:11:12,720 --> 00:11:16,320
So water and in the forecast, you can see.

133
00:11:17,350 --> 00:11:19,990
Does an array of three different durry?

134
00:11:20,730 --> 00:11:22,760
First, you have the forecasted value.

135
00:11:23,550 --> 00:11:25,770
Then you have the standard deviation.

136
00:11:26,370 --> 00:11:31,080
And then you have the 95 percent confidence interval for that value.

137
00:11:31,890 --> 00:11:37,470
So if you just want to select the forecasted value, you can write.

138
00:11:40,790 --> 00:11:42,110
Since you won the first.

139
00:11:43,220 --> 00:11:46,780
And this one because in the first set, you wanted this value.

140
00:11:46,880 --> 00:11:52,530
So if you run this, you will get the forecasted value, which is equal to this value.

141
00:11:53,240 --> 00:11:57,140
So 636 is the forecast set value for the next bidder.

142
00:11:57,920 --> 00:12:01,050
If you want to get these values for my people, put it.

143
00:12:01,130 --> 00:12:10,010
So suppose if I want to get values for next five period, I can just straight forecast and I can provide

144
00:12:10,160 --> 00:12:12,710
an integer value for the time bidders.

145
00:12:12,830 --> 00:12:14,470
I want the forecast for.

146
00:12:14,720 --> 00:12:20,630
So if I had a straight model under Sawford, Dorte forecast an argument is five.

147
00:12:22,040 --> 00:12:25,940
I will get the value for next five time periods.

148
00:12:26,420 --> 00:12:31,400
So in the first array we have the forecasted value in the second that a.

149
00:12:31,430 --> 00:12:31,630
B.

150
00:12:31,820 --> 00:12:33,920
The standard deviation.

151
00:12:34,190 --> 00:12:40,460
And in the last study I have the confidence interval for those forecasted values.

152
00:12:42,860 --> 00:12:50,430
Again, if I just want any particular value, I can use a square bracket to select that particular value.

153
00:12:52,790 --> 00:12:56,000
You can also use these values to plot the graph.

154
00:12:56,030 --> 00:13:00,410
So I just read this.

155
00:13:00,440 --> 00:13:05,300
I will get the first today and then I can plot this day on a later.

156
00:13:06,590 --> 00:13:09,780
So that's all for Atima in the next lecture.

157
00:13:09,800 --> 00:13:15,070
We will look at how to use walk forward validation for animal models.