1
00:00:00,209 --> 00:00:02,730
[MUSIC]

2
00:00:02,730 --> 00:00:05,663
Okay.
Well, we've talked quite extensively about

3
00:00:05,663 --> 00:00:09,535
using regression for the task of
predicting house value, but of course,

4
00:00:09,535 --> 00:00:14,370
the number of applications in which
regression can be used are quite large.

5
00:00:14,370 --> 00:00:16,426
Let's discuss just a few examples.

6
00:00:16,426 --> 00:00:17,166
So one is for

7
00:00:17,166 --> 00:00:22,126
predicting your salary after you take
this Machine Learning specialization.

8
00:00:22,126 --> 00:00:26,377
So you work really,
really hard in this specialization, and

9
00:00:26,377 --> 00:00:32,420
hopefully after that, you can go earn
a really cool job, have a very big salary.

10
00:00:32,420 --> 00:00:36,787
But to predict how much your salary will
be, maybe that depends on things like,

11
00:00:36,787 --> 00:00:40,844
what was your performance in the various
courses of the specialization?

12
00:00:40,844 --> 00:00:43,571
What was the quality of
your capstone project?

13
00:00:43,571 --> 00:00:47,010
How many forum responses did you make?

14
00:00:47,010 --> 00:00:48,200
And so on.

15
00:00:48,200 --> 00:00:50,564
Okay.
When we think about predicting your

16
00:00:50,564 --> 00:00:56,287
salary, y hat What we're gonna do is we're
gonna estimate our model parameters, w0,

17
00:00:56,287 --> 00:01:01,475
w1, w2 and w3, which are our intercept and
our weights on your performance in

18
00:01:01,475 --> 00:01:07,100
the class, the quality of your capstone,
and your participation in the forum.

19
00:01:07,100 --> 00:01:11,068
And when we're estimating these model
parameters, what we're gonna use,

20
00:01:11,068 --> 00:01:15,880
we're gonna use these features for other
students who have taken this course and

21
00:01:15,880 --> 00:01:18,940
the observations of what
their salaries were and

22
00:01:18,940 --> 00:01:23,290
the jobs they got after taking this
Machine Learning Specialization.

23
00:01:23,290 --> 00:01:27,120
Another application of regression is for
doing stock prediction.

24
00:01:27,120 --> 00:01:30,813
So here, we might think about
the fact that the prediction for

25
00:01:30,813 --> 00:01:35,516
the price of a stock tomorrow depends on
the recent history of the stock price.

26
00:01:35,516 --> 00:01:39,891
It depends on news events,
different things going on in the world,

27
00:01:39,891 --> 00:01:43,212
as well as the value of
other related commodities.

28
00:01:43,212 --> 00:01:47,385
And a very different application than
thinking about predicting stock prices or

29
00:01:47,385 --> 00:01:51,066
your salary after this specialization
is you're sitting on Twitter,

30
00:01:51,066 --> 00:01:52,416
you tweet something, and

31
00:01:52,416 --> 00:01:57,340
you wanna know how many people are gonna
eventually retweet what you just tweeted.

32
00:01:57,340 --> 00:02:01,997
Well, this might depend on how many
followers you have, how many followers

33
00:02:01,997 --> 00:02:06,582
your followers have, different features
of the text of what you tweeted,

34
00:02:06,582 --> 00:02:09,348
the popularity of
the hashtag you included,

35
00:02:09,348 --> 00:02:14,275
number of past retweets of your tweets
that you've had, and things like this.

36
00:02:14,275 --> 00:02:17,494
And it turns out these types of models
can actually be really good at predicting

37
00:02:17,494 --> 00:02:19,460
the eventual number of
retweets of your tweet.

38
00:02:21,300 --> 00:02:24,400
And another very different application
is maybe you have a smart house,

39
00:02:24,400 --> 00:02:26,980
there's lots of different
sensors in this house, and

40
00:02:26,980 --> 00:02:29,900
lots of different things you can control
like temperature settings in different

41
00:02:29,900 --> 00:02:34,230
locations, vents, window blinds,
and things like this.

42
00:02:34,230 --> 00:02:39,514
And you have a home office and you have
a desk, and you wanna be able to predict

43
00:02:39,514 --> 00:02:45,556
what's the temperature at your desk based
on various settings of all these controls.

44
00:02:45,556 --> 00:02:48,481
But there's actually no
sensor directly at your desk.

45
00:02:48,481 --> 00:02:51,240
So, what you can think about doing is

46
00:02:51,240 --> 00:02:56,228
fitting a spatial model in order
to predict what the temperature is

47
00:02:56,228 --> 00:03:00,519
at all these different
locations throughout the house.

48
00:03:00,519 --> 00:03:05,565
And this might depend on features such
as the thermostat settings in the house,

49
00:03:05,565 --> 00:03:10,311
whether the blinds are open or closed,
temperature outside, how you've

50
00:03:10,311 --> 00:03:15,078
set different vents, and as well as
things like the time of the day, okay?

51
00:03:15,078 --> 00:03:18,443
So, hopefully, you see that
applications of regression, and

52
00:03:18,443 --> 00:03:22,618
this really just touches the tip of the
iceberg of where regression is used, but

53
00:03:22,618 --> 00:03:24,574
it's really quite a powerful tool.

54
00:03:24,574 --> 00:03:28,679
[MUSIC]