1
00:00:00,000 --> 00:00:04,956
[MUSIC]

2
00:00:04,956 --> 00:00:08,473
When there's a trade off between
precision and recall, it's important for

3
00:00:08,473 --> 00:00:09,890
us to look at the two extremes.

4
00:00:09,890 --> 00:00:12,950
What does it mean to have a classifier
that's extremely precise?

5
00:00:12,950 --> 00:00:16,550
And what does it mean to have a classifier
that's extremely high recall?

6
00:00:16,550 --> 00:00:19,770
And how the two can go
against each other sometimes.

7
00:00:20,782 --> 00:00:24,300
First, let's think about what I
call an optimistic classifier.

8
00:00:24,300 --> 00:00:27,170
You might know some of these
optimists in your life.

9
00:00:27,170 --> 00:00:28,730
They think everything is good.

10
00:00:28,730 --> 00:00:29,870
How's it going?

11
00:00:29,870 --> 00:00:31,436
Good.
Even if bad stuff is happening,

12
00:00:31,436 --> 00:00:32,110
they say good.

13
00:00:33,480 --> 00:00:37,420
Those folks say that all
possible experiences are good, so

14
00:00:37,420 --> 00:00:38,540
they're optimists.

15
00:00:38,540 --> 00:00:42,450
That means that pretty much every input,
every sentence,

16
00:00:42,450 --> 00:00:45,780
is labeled as positive,
very few get labeled as negative.

17
00:00:45,780 --> 00:00:53,050
It's extremely likely that all the truly
positive data points get labeled as good.

18
00:00:53,050 --> 00:00:54,270
What does that mean?

19
00:00:54,270 --> 00:00:58,370
That means that I have perfect recall,

20
00:00:58,370 --> 00:01:01,150
because I recall all those
positive data points.

21
00:01:01,150 --> 00:01:02,110
Good.

22
00:01:02,110 --> 00:01:05,180
But I might not get

23
00:01:05,180 --> 00:01:08,460
perfect precision because I put in
a bunch of negatives into that bit.

24
00:01:09,690 --> 00:01:11,290
How can we address that?

25
00:01:11,290 --> 00:01:13,830
We can have that pessimistic classifier,

26
00:01:13,830 --> 00:01:16,940
you might have some friends like
that where you try really hard,

27
00:01:16,940 --> 00:01:22,350
you do everything for them, you go out
of your way, and everything sucks.

28
00:01:22,350 --> 00:01:25,460
Every single experience that
you have is really bad.

29
00:01:25,460 --> 00:01:29,490
There's very, very, very, very,
few things that they say are good.

30
00:01:29,490 --> 00:01:34,902
And when they are there very likely to be
good but everything else they say is bad,

31
00:01:34,902 --> 00:01:38,912
so everything else in the world
is very hard, equals -1.

32
00:01:38,912 --> 00:01:42,370
Pessimist means that you're going to
miss out on many good things in life.

33
00:01:43,900 --> 00:01:48,940
The pessimists have high precision because
the few things that was good tends to be

34
00:01:48,940 --> 00:01:55,270
good, but very, very, very low recall,
they don't inspire great things in life.

35
00:01:56,620 --> 00:02:02,200
It turns out there is a spectrum between
a high precision low recall model and

36
00:02:02,200 --> 00:02:04,770
low precision high recall model,
the pessimist and the optimist.

37
00:02:06,350 --> 00:02:10,690
What we'd like to do is somehow balance
between the two perspectives in the world

38
00:02:10,690 --> 00:02:13,810
to find something that's just right for
us.

39
00:02:13,810 --> 00:02:18,610
So, balance between a pessimistic
model and the optimistic model.

40
00:02:20,510 --> 00:02:25,920
In particular, we want to find as many
positive reviews or sentences as possible,

41
00:02:25,920 --> 00:02:31,420
as many of those as possible, with as
few incorrect predictions as we can.

42
00:02:31,420 --> 00:02:35,272
So, that's the balance we're trying to
strike in the case of our restaurant.

43
00:02:35,272 --> 00:02:39,309
[MUSIC]