1
00:00:00,000 --> 00:00:03,180
Welcome to the second week
of this course.

2
00:00:03,180 --> 00:00:04,770
You learned about how to compute

3
00:00:04,770 --> 00:00:07,140
statistical analysis
of a time series.

4
00:00:07,140 --> 00:00:09,165
In this week, you'll
learn to apply

5
00:00:09,165 --> 00:00:11,430
a new network to these sequences.

6
00:00:11,430 --> 00:00:14,420
Yeah, we'll start with
a relatively simple DNN,

7
00:00:14,420 --> 00:00:16,020
and if you remember
DNNs all the way

8
00:00:16,020 --> 00:00:18,000
back like at the beginning
of the specialization,

9
00:00:18,000 --> 00:00:19,770
so you're going to be able
to bring those skills

10
00:00:19,770 --> 00:00:21,735
to bear on time-series data.

11
00:00:21,735 --> 00:00:23,790
So we're going to build
a very simple DNN

12
00:00:23,790 --> 00:00:25,425
just like a three-layer DNN.

13
00:00:25,425 --> 00:00:27,870
Then one technique
though that's I

14
00:00:27,870 --> 00:00:30,480
find really useful that you're
going to learn this week,

15
00:00:30,480 --> 00:00:33,570
is being able to tune the
learning rate of the optimizer.

16
00:00:33,570 --> 00:00:36,545
We'll all spent a lot of time
hand-tuning learning rate.

17
00:00:36,545 --> 00:00:38,960
One of my most distinctive
memories from couple of

18
00:00:38,960 --> 00:00:41,570
decades ago was
a summer intern at

19
00:00:41,570 --> 00:00:43,970
Microsoft research
staying up until

20
00:00:43,970 --> 00:00:46,925
3:00 AM in the office,
hand-tuning learning rates.

21
00:00:46,925 --> 00:00:49,475
I wish I could tune
my own learning rate sometimes.

22
00:00:49,475 --> 00:00:52,955
So I think giving people
a systematic way to do this,

23
00:00:52,955 --> 00:00:54,430
will be very useful.

24
00:00:54,430 --> 00:00:55,700
Yeah. I think so,

25
00:00:55,700 --> 00:00:58,040
and I know I've spent
too many sleepless nights at it.

26
00:00:58,040 --> 00:01:00,230
So we tried to automate
it here and so teach you

27
00:01:00,230 --> 00:01:01,310
a technique for automating

28
00:01:01,310 --> 00:01:02,960
it and hopefully it'll be useful,

29
00:01:02,960 --> 00:01:05,000
and hopefully it'll help
you to start learning

30
00:01:05,000 --> 00:01:07,525
how to predict
your time series data.

31
00:01:07,525 --> 00:01:11,390
By the end of this week, you
have trained DNN on time

32
00:01:11,390 --> 00:01:13,130
series data that
you saw last week

33
00:01:13,130 --> 00:01:15,050
and you get to see
how well it does.

34
00:01:15,050 --> 00:01:17,250
Please go onto to the next video.