1
00:00:00,433 --> 00:00:01,300
Hello and welcome back.

2
00:00:01,300 --> 00:00:03,333
Today we're talking about the importance
of splitting

3
00:00:03,333 --> 00:00:06,333
your data set into a training set
and a test set.

4
00:00:06,333 --> 00:00:11,266
Let's imagine that you are tasked
to predict the sale prices of cars.

5
00:00:11,533 --> 00:00:14,233
And that is your dependent variable.

6
00:00:14,233 --> 00:00:18,466
And your independent variables
are the mileage of the car.

7
00:00:18,466 --> 00:00:20,766
And its age.

8
00:00:20,766 --> 00:00:25,666
And in your, data that was supplied
to you, you have 20 cars in total.

9
00:00:25,700 --> 00:00:28,700
Of course, that's not a lot,
but for illustrative purposes,

10
00:00:28,800 --> 00:00:31,333
for our tutorial, that will be sufficient.

11
00:00:31,333 --> 00:00:33,900
So what splitting your data implies

12
00:00:33,900 --> 00:00:36,900
is, separating a part of your data out.

13
00:00:37,200 --> 00:00:39,133
before you do anything.

14
00:00:39,133 --> 00:00:41,300
And usually that's about 20% of the data.

15
00:00:41,300 --> 00:00:46,600
So since we have, 20 cars here, that's
about four cars that we separate out.

16
00:00:46,733 --> 00:00:50,166
So what that means is that the bulk of
our data, 80%, will be our training set,

17
00:00:50,466 --> 00:00:53,666
and the separated 20%
will be our test set.

18
00:00:54,000 --> 00:00:57,533
We'll use our training set
to build the model.

19
00:00:57,733 --> 00:01:00,733
So in this case
we're building a linear regression.

20
00:01:01,100 --> 00:01:05,433
And then we will take the cars
from the test set.

21
00:01:05,733 --> 00:01:08,300
We will apply our model to them.

22
00:01:08,300 --> 00:01:12,300
So they haven't been part of the model
creation process.

23
00:01:12,300 --> 00:01:14,533
The model has no information
about these cars.

24
00:01:14,533 --> 00:01:16,866
And now we're applying this model to them.

25
00:01:16,866 --> 00:01:21,233
And it's predicting
certain, values, certain prices.

26
00:01:21,433 --> 00:01:25,800
But the good news is that because
this is something we separated in advance

27
00:01:25,800 --> 00:01:29,966
as part of the data that was given to us,
we actually know the actual prices.

28
00:01:30,066 --> 00:01:33,933
So now we can compare
the predicted values,

29
00:01:34,166 --> 00:01:39,366
which were generated using a model
that has never seen these cars before.

30
00:01:39,900 --> 00:01:43,833
And we can compare
that to the actual values that we know

31
00:01:43,833 --> 00:01:45,900
what these cars sold for.

32
00:01:45,900 --> 00:01:48,900
And so from that we can evaluate
our model.

33
00:01:48,900 --> 00:01:50,100
Is it doing a good job.

34
00:01:50,100 --> 00:01:52,133
Is it doing a not so good job.

35
00:01:52,133 --> 00:01:53,633
And do we need to improve it.

36
00:01:53,633 --> 00:01:56,033
And that's how we split a training set
and a test set.

37
00:01:56,033 --> 00:01:58,033
And that's why it's important to do that.

38
00:01:58,033 --> 00:01:59,766
I look forward to seeing you
in the next tutorial.

39
00:01:59,766 --> 00:02:01,666
And until then, enjoy machine learning.