1
00:00:00,040 --> 00:00:04,912
[MUSIC]

2
00:00:04,912 --> 00:00:09,543
As we've seen, overfitting decision trees
is a fundamental problem that must be

3
00:00:09,543 --> 00:00:10,770
addressed.

4
00:00:10,770 --> 00:00:14,000
And we've looked at a couple different
ways to address this problem.

5
00:00:14,000 --> 00:00:16,140
We looked at early stopping.

6
00:00:16,140 --> 00:00:17,310
So as we go down the tree,

7
00:00:17,310 --> 00:00:22,600
we're going to stop early to avoid getting
to a very complex tree and overfitting.

8
00:00:22,600 --> 00:00:26,100
And then we talked about pruning, where
we take a massive tree and then we chop,

9
00:00:26,100 --> 00:00:28,480
chop, chop, chop and simplify it.

10
00:00:28,480 --> 00:00:32,570
Pruning is very important in
practical settings, and so

11
00:00:32,570 --> 00:00:35,170
we spent quite a of bit of time going
through the details of that but

12
00:00:35,170 --> 00:00:37,850
it was optional section in the lecture.

13
00:00:37,850 --> 00:00:39,880
But I just want to make sure,

14
00:00:39,880 --> 00:00:42,380
we're all aware that if you
see an algorithm out there for

15
00:00:42,380 --> 00:00:47,130
decision trees that does pruning, that's
a kind of idea of what's being executed.

16
00:00:47,130 --> 00:00:51,660
So overfitting decision trees
is a fundamental topic, but

17
00:00:51,660 --> 00:00:54,190
one that now we have
a handle on how to address.

18
00:00:55,970 --> 00:01:00,112
And let me close by thanking my colleague,
Krishna Sridhar,

19
00:01:00,112 --> 00:01:04,176
who has been really fundamental
at creating these slides and

20
00:01:04,176 --> 00:01:07,777
being able to make this
lecture as great as possible.

21
00:01:07,777 --> 00:01:12,929
[MUSIC]