1
00:00:00,000 --> 00:00:03,900
Note that when we use the
trailing window when computing

2
00:00:03,900 --> 00:00:06,015
the moving average
of present values

3
00:00:06,015 --> 00:00:08,460
from t minus 32, t minus one.

4
00:00:08,460 --> 00:00:10,770
But when we use
a centered window to compute

5
00:00:10,770 --> 00:00:14,640
the moving average of
past values from one year ago,

6
00:00:14,640 --> 00:00:17,670
that's t minus one year
minus five days,

7
00:00:17,670 --> 00:00:20,370
to t minus one year
plus five days.

8
00:00:20,370 --> 00:00:23,105
Then moving averages
using centered windows

9
00:00:23,105 --> 00:00:26,105
can be more accurate than
using trailing windows.

10
00:00:26,105 --> 00:00:29,075
But we can't use
centered windows to smooth

11
00:00:29,075 --> 00:00:32,645
present values since we
don't know future values.

12
00:00:32,645 --> 00:00:35,135
However, to smooth past values

13
00:00:35,135 --> 00:00:38,230
we can afford to use
centered windows.

14
00:00:38,230 --> 00:00:40,550
Okay, so now we've looked at

15
00:00:40,550 --> 00:00:42,110
a few statistical methods for

16
00:00:42,110 --> 00:00:44,705
predicting the next values
in a time series.

17
00:00:44,705 --> 00:00:46,760
In the next video,
you'll take a look at

18
00:00:46,760 --> 00:00:49,355
a screencast of
this prediction in action.

19
00:00:49,355 --> 00:00:52,095
Once you've done
the statistical forecasting,

20
00:00:52,095 --> 00:00:53,690
the next step of
course will be to

21
00:00:53,690 --> 00:00:55,460
apply the machine-learning
techniques

22
00:00:55,460 --> 00:00:56,780
you've been learning all along in

23
00:00:56,780 --> 00:00:59,760
TensorFlow and you'll
do that next week.