WEBVTT - This file was automatically generated by VIMEO

0
00:00:07.760 --> 00:00:08.240
Hi.

1
00:00:08.240 --> 00:00:12.820
In this video we will learn a little bit about the XG Boost algorithm.

2
00:00:15.190 --> 00:00:17.370
XG Boost is a very, uh,

3
00:00:17.370 --> 00:00:21.250
popular machine learning algorithm I use for classification,

4
00:00:21.540 --> 00:00:24.340
regression and ranking problems. Uh,

5
00:00:24.360 --> 00:00:27.780
it stands for Extreme Gradient Boosting. Uh,

6
00:00:27.890 --> 00:00:30.540
it's very scalable and fast. Uh,

7
00:00:30.720 --> 00:00:33.580
it is an ensemble learning method, uh,

8
00:00:33.610 --> 00:00:38.500
that means that multiple weak decision trees are trained and their

9
00:00:38.500 --> 00:00:43.440
predictions are used to create a strong final prediction. It has,

10
00:00:43.500 --> 00:00:46.360
uh, support for parallel processing, uh,

11
00:00:46.380 --> 00:00:49.560
so we can train g boost on multiple cores and machines.

12
00:00:51.240 --> 00:00:55.140
It also supports outof core computation, uh, which means, uh,

13
00:00:55.160 --> 00:00:59.300
it can perform training on data sets that xts available ram.

14
00:01:00.000 --> 00:01:03.940
It also allows training on GPUs. Uh,

15
00:01:04.040 --> 00:01:08.500
it has a built-in support for handling missing values in the input dataset.

16
00:01:09.800 --> 00:01:14.740
And, uh, while EG Boost does not explicitly remove, uh, variables,

17
00:01:15.410 --> 00:01:16.540
it's process, uh,

18
00:01:16.560 --> 00:01:21.460
of building decision trees involves selecting features that contribute

19
00:01:21.600 --> 00:01:26.160
the most to reducing prediction errors. Exg boot, uh,

20
00:01:26.340 --> 00:01:29.840
has uh, no assumptions on input data distribution.

21
00:01:31.190 --> 00:01:34.210
It also doesn't have any assumption that, uh,

22
00:01:34.230 --> 00:01:39.130
the input features and target are in linear relationship.

23
00:01:40.790 --> 00:01:45.410
And, uh, although outliers, uh, can affect model performance,

24
00:01:46.150 --> 00:01:50.610
uh, XG Boost three base nature makes it, uh, robust, uh,

25
00:01:50.610 --> 00:01:54.350
against them into some extent. Uh, decision trees, uh,

26
00:01:54.350 --> 00:01:58.310
naturally adapt to outliers by creating splits that, uh,

27
00:01:58.670 --> 00:02:03.540
minimize the impact of extreme values. In our next video,

28
00:02:03.920 --> 00:02:06.580
we will create our binary XG Boost model.
