1
00:00:00,000 --> 00:00:04,914
[MUSIC]

2
00:00:04,914 --> 00:00:08,446
In this module, we will discuss some
very important fundamental concept,

3
00:00:08,446 --> 00:00:10,390
which is evaluating classifiers.

4
00:00:10,390 --> 00:00:14,165
And in particular, we talked about
precision-recall, which is a concept

5
00:00:14,165 --> 00:00:18,535
that's widely used, way beyond some of
the classifiers we talked about here.

6
00:00:18,535 --> 00:00:21,975
So basically any classification problem
you're going to see in industry.

7
00:00:21,975 --> 00:00:24,115
We saw that just straight accuracy or

8
00:00:24,115 --> 00:00:26,694
error metrics may not be the right
things for your application.

9
00:00:26,694 --> 00:00:29,582
And you need to look at something else and

10
00:00:29,582 --> 00:00:33,562
precision recall is one of the first
things you might want to look at.

11
00:00:33,562 --> 00:00:36,282
Precision captures the fraction

12
00:00:36,282 --> 00:00:40,012
of your top of your positive predictions
that are actually positive and

13
00:00:40,012 --> 00:00:45,772
recall talks about of all positive
predictions positive sentences out there,

14
00:00:45,772 --> 00:00:48,640
which ones did you find,
which ones did you label as positive.

15
00:00:48,640 --> 00:00:51,750
So we talked about the trade-off
in between precision and recall,

16
00:00:51,750 --> 00:00:56,490
and how you can navigate that trade-off
with that trade-off parameter t,

17
00:00:56,490 --> 00:01:01,270
in terms of probability, and really get
this beautiful precision trade-off curves.

18
00:01:01,270 --> 00:01:06,870
And finally, we talked about comparing
models with this precision at k metric,

19
00:01:06,870 --> 00:01:09,409
which is one that I particularly like for
a lot of applications.

20
00:01:10,450 --> 00:01:14,816
I want to take a moment here to
thank my colleague, Krishna Sridhar,

21
00:01:14,816 --> 00:01:18,801
who is instrumental at helping
these slides come together and

22
00:01:18,801 --> 00:01:21,271
creating the ideas behind this module

23
00:01:21,271 --> 00:01:25,909
[MUSIC]