1
00:00:00,480 --> 00:00:05,600
Now we are going to talk about simple exponential smoothing in moving.

2
00:00:05,710 --> 00:00:10,590
This morning, we were taking the average of some of the values of the cities.

3
00:00:12,150 --> 00:00:18,330
But when we do simple exponential smoothing, we do a weighted average of the values.

4
00:00:20,250 --> 00:00:29,730
The concept is that the importance of latest value will be sometimes more than the importance of order

5
00:00:29,730 --> 00:00:30,270
values.

6
00:00:31,860 --> 00:00:38,250
So we will assign larger weights to the latest values and smaller weights to the older ones.

7
00:00:39,850 --> 00:00:41,740
Here is the formula for the same.

8
00:00:45,330 --> 00:00:51,300
The new forecast where Lou is given as Alpha Times last value.

9
00:00:51,660 --> 00:00:56,310
Plus, I'll find into one minus Alpha times last to last value and so on.

10
00:00:58,290 --> 00:01:05,790
Hit Alpha is a constant between zero and one, and it is called smoothing constant.

11
00:01:07,560 --> 00:01:14,880
Remember that we can only use moving average smoothing and exponential spudding when there is no brain

12
00:01:15,030 --> 00:01:16,740
or seasonality in the city's.

13
00:01:19,600 --> 00:01:22,480
The value of Alpha is to be selected by us.

14
00:01:23,170 --> 00:01:29,650
If we keep Alpha close to one, then we give more importance to the decent values of the city's.

15
00:01:30,920 --> 00:01:36,880
Such a model is called a fast learner because it adapt quickly with the reset values.

16
00:01:38,690 --> 00:01:44,560
If you keep ULFA close to zero, then we give more importance to older values.

17
00:01:45,740 --> 00:01:47,870
Such a model is called a slow learner.

18
00:01:50,600 --> 00:01:55,640
So the choice of elford depends on the amount of smoothing and the relevance of history.

19
00:01:55,850 --> 00:02:03,410
We want to maintain in our model when we are using a socket, we can test on multiple values of alpha

20
00:02:03,950 --> 00:02:06,500
and keep the one which gives minimum error.