1
00:00:00,000 --> 00:00:04,281
[MUSIC]

2
00:00:04,281 --> 00:00:08,324
Well, the last thing that I, I want to
cover in this module is the fact that

3
00:00:08,324 --> 00:00:12,710
we've looked at a very simple notion of
errors, this residual sum of squares.

4
00:00:12,710 --> 00:00:16,320
And it's actually really,
really commonly used in practice.

5
00:00:16,320 --> 00:00:19,070
But we're gonna talk a lot more about

6
00:00:19,070 --> 00:00:21,480
different types of errors
later in the course.

7
00:00:21,480 --> 00:00:24,710
But I wanted to go through some of
the intuition of what happens if we

8
00:00:24,710 --> 00:00:26,850
use a different measure of air.

9
00:00:26,850 --> 00:00:30,200
Okay so this residual sum of
squares that we've been looking at

10
00:00:30,200 --> 00:00:32,500
Is something that's called
a symmetric cost function.

11
00:00:32,500 --> 00:00:37,310
And that's because what we're assuming
when we look at this aerometric

12
00:00:37,310 --> 00:00:42,240
is the fact that if we over
estimate the value of our house,

13
00:00:42,240 --> 00:00:46,820
that has the same cost as if we under
estimate the value of the house.

14
00:00:46,820 --> 00:00:49,890
And the question is so
we actually believe that is the case?

15
00:00:49,890 --> 00:00:54,990
And what happens if there might not
be symmetric cost to these error?

16
00:00:54,990 --> 00:00:59,800
But what if the cost of
listing my house sales price

17
00:00:59,800 --> 00:01:04,740
as too high is bigger than the cost
if I listed it as too low?

18
00:01:04,740 --> 00:01:09,580
So for example,
if I list the value as too high,

19
00:01:09,580 --> 00:01:11,890
then maybe no one will
even come see the house.

20
00:01:11,890 --> 00:01:14,230
Or they come see it and
they say oh it's definitely not worth it.

21
00:01:14,230 --> 00:01:16,400
I'm not putting in an offer.

22
00:01:16,400 --> 00:01:18,870
So the result might be I get no offers.

23
00:01:18,870 --> 00:01:21,170
So as a seller, that's a big cost to me.

24
00:01:21,170 --> 00:01:23,650
I've gone through this whole
thing of trying to sell my house.

25
00:01:23,650 --> 00:01:26,110
I want or need to sell my house.

26
00:01:26,110 --> 00:01:28,100
And I get no offers.

27
00:01:28,100 --> 00:01:29,230
On the other hand,

28
00:01:29,230 --> 00:01:34,480
if I list the sales price as too low,
of course I won't get offers as high

29
00:01:34,480 --> 00:01:39,650
as I could have if I had more accurately
estimated the value of the house.

30
00:01:39,650 --> 00:01:41,600
But I still get offers, and

31
00:01:41,600 --> 00:01:46,410
maybe that cost to me Is less bad
than getting no offers at all.

32
00:01:46,410 --> 00:01:51,010
For example if I have to move to
another state I have no choice but

33
00:01:51,010 --> 00:01:52,840
to sell my house.

34
00:01:52,840 --> 00:01:53,380
Okay, so

35
00:01:53,380 --> 00:01:58,610
in this case it might be more appropriate
to use an asymmetric cost function

36
00:01:58,610 --> 00:02:04,930
where the errors are not weighed equally
between these two types of mistakes.

37
00:02:04,930 --> 00:02:09,000
And instead of this dash orange line here,

38
00:02:13,100 --> 00:02:19,630
which represents our fit when we're
minimizing residual sum of squares.

39
00:02:19,630 --> 00:02:21,930
I will get some different solution.

40
00:02:22,990 --> 00:02:25,849
Again, sorry,
I love to write over my animations.

41
00:02:27,390 --> 00:02:32,095
Here this is the fit minimizing
residual sum of squares, and

42
00:02:32,095 --> 00:02:38,103
this other orange line here is this
other solution using an asymmetric loss.

43
00:02:46,299 --> 00:02:53,785
Asymmetric Loss and
specifically in asymmetric loss.

44
00:02:53,785 --> 00:02:55,965
Let me just say asymmetric cost.

45
00:03:00,063 --> 00:03:05,500
Where I prefer to underestimate
the value than over.

46
00:03:05,500 --> 00:03:08,280
And that's what you see here is,
in general,

47
00:03:08,280 --> 00:03:10,990
we're predicting the values as lower.

48
00:03:10,990 --> 00:03:16,630
As compared to the line that we got or our
predictions using residual sum of squares.

49
00:03:18,170 --> 00:03:22,987
Okay so that's just a little bit of
intuition about what would happen using

50
00:03:22,987 --> 00:03:24,900
different cost functions and

51
00:03:24,900 --> 00:03:29,348
again we're gonna talk a lot more
about this later on in this course.

52
00:03:29,348 --> 00:03:33,279
[MUSIC]