1
00:00:04,500 --> 00:00:08,310
In this video we'll try to make
predictions for the 2012-2013

2
00:00:08,310 --> 00:00:09,560
season.

3
00:00:09,560 --> 00:00:13,020
We'll need to load our test set
because our training set only

4
00:00:13,020 --> 00:00:18,870
included data from 1980 up
until the 2011-2012 season.

5
00:00:18,870 --> 00:00:19,960
So let's call it NBA_test.

6
00:00:25,470 --> 00:00:27,550
And we'll read it in
the same way as always,

7
00:00:27,550 --> 00:00:28,600
read.csv("NBA_test.csv").

8
00:00:36,220 --> 00:00:40,160
All right, so now let's try
to predict using our model

9
00:00:40,160 --> 00:00:43,090
that we made in the previous
video, how many points we'll

10
00:00:43,090 --> 00:00:46,070
see in 2012-2013 season.

11
00:00:46,070 --> 00:00:47,450
Let's call this
PointsPrediction.

12
00:00:54,120 --> 00:00:57,450
And so we use the
predict command here.

13
00:00:57,450 --> 00:00:59,990
And we give it the previous
model that we made.

14
00:01:02,770 --> 00:01:05,620
We'll give it
PointsReg4, because that

15
00:01:05,620 --> 00:01:09,030
was the model we determined
at the end to be the best one.

16
00:01:09,030 --> 00:01:11,880
And the new data
which is NBA_test.

17
00:01:16,150 --> 00:01:21,430
OK, so now that we have our
prediction, how good is it?

18
00:01:21,430 --> 00:01:24,570
We can compute the out
of sample r-squared.

19
00:01:24,570 --> 00:01:25,980
This is a measurement
of how well

20
00:01:25,980 --> 00:01:28,580
the model predicts on test data.

21
00:01:28,580 --> 00:01:31,200
The r squared value we
had before from our model,

22
00:01:31,200 --> 00:01:35,300
the 0.8991, you might
remember, is the measure

23
00:01:35,300 --> 00:01:37,610
of an in-sample
r-squared, which is

24
00:01:37,610 --> 00:01:40,580
how well the model
fits the training data.

25
00:01:40,580 --> 00:01:43,490
But to get a measure of the
predictions goodness of fit,

26
00:01:43,490 --> 00:01:46,740
we need to calculate the
out of sample r-squared.

27
00:01:46,740 --> 00:01:48,310
So let's do that here.

28
00:01:48,310 --> 00:01:50,190
We need to compute the
sum of squared errors.

29
00:01:55,240 --> 00:02:02,370
And so this here is just the
sum of the predicted amount

30
00:02:02,370 --> 00:02:11,470
minus the actual amount of
points squared and summed.

31
00:02:11,470 --> 00:02:15,670
And we need the total
sums of squares,

32
00:02:15,670 --> 00:02:24,890
which is just the sum of
the average number of points

33
00:02:24,890 --> 00:02:27,680
minus the test actual
number of points.

34
00:02:38,380 --> 00:02:41,530
So the r-squared here
then is calculated

35
00:02:41,530 --> 00:02:45,630
as usual, 1 minus the sum
of squared errors divided

36
00:02:45,630 --> 00:02:48,500
by total sums of squares.

37
00:02:48,500 --> 00:02:53,100
And we see that we have an
r squared value of 0.8127.

38
00:02:53,100 --> 00:02:56,810
We can also calculate the root
mean square error the same way

39
00:02:56,810 --> 00:02:58,940
as before, root
mean squared error

40
00:02:58,940 --> 00:03:03,080
is going to be the square root
of the sum of squared errors

41
00:03:03,080 --> 00:03:06,140
divided by n, which is the
number of rows in our test data

42
00:03:06,140 --> 00:03:06,640
set.

43
00:03:13,810 --> 00:03:21,280
OK and the root mean squared
error here is 196.37.

44
00:03:21,280 --> 00:03:23,120
So it's a little bit
higher than before.

45
00:03:23,120 --> 00:03:24,400
But it's not too bad.

46
00:03:24,400 --> 00:03:29,680
We're making an average
error of about 196 points.

47
00:03:29,680 --> 00:03:31,220
We'll stop here for now.

48
00:03:31,220 --> 00:03:33,220
Good luck with the homework.