1
00:00:00,410 --> 00:00:02,640
Out of interest, let's do a prediction.

2
00:00:02,640 --> 00:00:09,180
The window size I'm using is 30 steps,
and the dataset is 3,235 steps long.

3
00:00:09,180 --> 00:00:13,554
So if I want to predict the next
value after the end of my dataset,

4
00:00:13,554 --> 00:00:15,099
I would use this code.

5
00:00:15,099 --> 00:00:19,275
And I would get the result 7.0773993.

6
00:00:20,756 --> 00:00:23,850
The dataset goes up to July 2018, so

7
00:00:23,850 --> 00:00:29,227
I'm actually predicting 7.077 sunspots for
August 2018.

8
00:00:29,227 --> 00:00:31,818
And if I look at this
chart of observations,

9
00:00:31,818 --> 00:00:35,529
which does have some slightly
different data from my dataset,

10
00:00:35,529 --> 00:00:40,540
I can see that the actual recorded number
of sunspots in August 2018 was 8.7.

11
00:00:40,540 --> 00:00:44,180
So the prediction isn't too bad, but
let's see if we can improve on it.

12
00:00:45,840 --> 00:00:50,369
With these settings,
I got the MAE down to 13.75, and

13
00:00:50,369 --> 00:00:57,177
the prediction was 8.13, which is much
closer to the actual real reading of 8.7.

14
00:00:57,177 --> 00:01:01,950
There is a random elements in creating
models, however, so your results may vary.

15
00:01:01,950 --> 00:01:05,400
Doing accuracy based on a single
prediction like this is also a recipe for

16
00:01:05,400 --> 00:01:06,730
disappointment, and

17
00:01:06,730 --> 00:01:10,250
you're much better off evaluating mean
accuracy over a number of readings.

18
00:01:11,790 --> 00:01:16,260
So here, we looked at using
a DNN to predict sunspot values.

19
00:01:16,260 --> 00:01:19,402
With a little bit of tuning,
we reduced our MAE a bit.

20
00:01:19,402 --> 00:01:23,280
And when we tried to predict the next
month's value using this model,

21
00:01:23,280 --> 00:01:25,456
we got quite close to the actual value.

22
00:01:25,456 --> 00:01:28,669
In the next video, you'll go through
this workbook to see it in action.

23
00:01:28,669 --> 00:01:30,657
And then you should try it out for
yourself,

24
00:01:30,657 --> 00:01:33,310
to see if you can improve on what I got.

25
00:01:33,310 --> 00:01:35,530
After that,
you'll look into the RNNs again and

26
00:01:35,530 --> 00:01:37,880
see if you can get better
predictions using those.