1
00:00:00,000 --> 00:00:04,789
[MUSIC]

2
00:00:04,789 --> 00:00:08,344
It may seem like more data is always
better, but we want to make sure

3
00:00:08,344 --> 00:00:12,940
we're only looking at information
that's relevant to our problem.

4
00:00:12,940 --> 00:00:15,900
Let's talk about some reasons you may
actually want to make your data set

5
00:00:15,900 --> 00:00:19,000
smaller by removing some of the features.

6
00:00:19,000 --> 00:00:20,210
When looking at data,

7
00:00:20,210 --> 00:00:24,290
we're usually looking to find
relationships between our features.

8
00:00:24,290 --> 00:00:27,520
Often, there's one feature we're
particularly interested in,

9
00:00:27,520 --> 00:00:30,570
sometimes called the target,
or dependant variable.

10
00:00:30,570 --> 00:00:34,380
This is the information we want
to predict, or learn more about.

11
00:00:34,380 --> 00:00:37,970
The other features are treated
as independent variables.

12
00:00:37,970 --> 00:00:42,480
We want to know what influence these
features have on our target variable.

13
00:00:42,480 --> 00:00:45,990
Including irrelevant data can
waste computing resources.

14
00:00:45,990 --> 00:00:49,110
Storing data takes up space
on a computer hard drive.

15
00:00:49,110 --> 00:00:52,700
So keeping data we don't
need wastes storage space.

16
00:00:52,700 --> 00:00:56,650
It also takes longer to run
an analysis on a larger data set.

17
00:00:56,650 --> 00:01:00,630
If some of that data isn't needed,
it makes your computer take extra time

18
00:01:00,630 --> 00:01:04,000
to process data that isn't
helping your analysis.

19
00:01:04,000 --> 00:01:06,200
It can also mislead an analysis,

20
00:01:06,200 --> 00:01:10,082
causing it to try to find trends
in data that don't actually exist.

21
00:01:10,082 --> 00:01:14,510
Large amounts of data are difficult
to visualize in a meaningful way.

22
00:01:14,510 --> 00:01:15,950
The more features we have,

23
00:01:15,950 --> 00:01:20,300
the more difficult it is to make
a clear visualization of that data.

24
00:01:20,300 --> 00:01:21,370
With two features,

25
00:01:21,370 --> 00:01:25,480
we can plot the relationship as
a simple scatter plot or line.

26
00:01:25,480 --> 00:01:27,190
As the number of features increases,

27
00:01:27,190 --> 00:01:30,230
it becomes more difficult
to visualize the data.

28
00:01:30,230 --> 00:01:34,850
We start to need 3D graphs or
complex pairings of 2D graphs.

29
00:01:34,850 --> 00:01:36,900
These make it more
difficult to look at and

30
00:01:36,900 --> 00:01:39,900
understand the relationships
within our data set.

31
00:01:39,900 --> 00:01:42,950
Getting rid of the unnecessary
features helps to focus on

32
00:01:42,950 --> 00:01:44,850
only the important information.

33
00:01:44,850 --> 00:01:47,310
And allows us to make a clearer graph.

34
00:01:47,310 --> 00:01:51,300
To address this, we want to choose
the most relevant data to explore or

35
00:01:51,300 --> 00:01:52,920
analyze further.

36
00:01:52,920 --> 00:01:55,530
We don't want to get rid
of any important data, but

37
00:01:55,530 --> 00:01:58,910
keeping only useful data
can improve our analysis.

38
00:01:58,910 --> 00:02:02,660
We can make our data sets smaller
by removing data we don't want or

39
00:02:02,660 --> 00:02:06,270
by grouping similar data
together into a single feature.

40
00:02:06,270 --> 00:02:07,770
In the next few videos,

41
00:02:07,770 --> 00:02:11,100
we'll talk about ways we can make sure
we're keeping only the best data.