0
1
00:00:00,330 --> 00:00:06,090
Okay so we've spec'd out our model and ran our first regression.
1

2
00:00:06,090 --> 00:00:12,450
We did a quick sense check of what we got back and we looked at the performance of our model.
2

3
00:00:12,450 --> 00:00:20,550
Now we're getting to our model evaluation stage. Evaluating and deploying the model is the final part
3

4
00:00:20,670 --> 00:00:22,060
of our workflow.
4

5
00:00:22,080 --> 00:00:24,480
We've done a lot of legwork up till now.
5

6
00:00:24,480 --> 00:00:26,380
We formulated our question.
6

7
00:00:26,430 --> 00:00:27,840
We've gathered our data.
7

8
00:00:27,930 --> 00:00:29,460
We've cleaned our data.
8

9
00:00:29,460 --> 00:00:33,190
We've explored and visualized our data, made sense of it.
9

10
00:00:33,210 --> 00:00:40,260
We've trained our algorithm and now we're getting to the point where we can evaluate our results, refine
10

11
00:00:40,260 --> 00:00:46,110
our model, look for problems, look for improvements and deploy our model.
11

12
00:00:46,440 --> 00:00:49,400
So what kind of things do we do at this stage?
12

13
00:00:49,710 --> 00:00:51,660
Where do we take it from here?
13

14
00:00:52,410 --> 00:00:58,900
Well we're going to start by asking a series of questions about our model and about our results.
14

15
00:00:58,980 --> 00:01:05,130
We're going to be checking for some problems and looking for improvements retraining our algorithm as
15

16
00:01:05,130 --> 00:01:06,300
necessary.
16

17
00:01:06,300 --> 00:01:11,430
The approach that we're going to take is kind of like going for a medical checkup.
17

18
00:01:11,430 --> 00:01:16,030
When you go see a doctor the doctor's typically ran some tests on you.
18

19
00:01:16,030 --> 00:01:16,520
Right?
19

20
00:01:16,560 --> 00:01:22,200
They measure your height, your weight, listen to your breathing, take an X-ray, measure your blood pressure,
20

21
00:01:22,440 --> 00:01:23,880
measure your heart rate,
21

22
00:01:23,880 --> 00:01:28,170
run a blood test, hit your knee with a tiny little hammer and so on.
22

23
00:01:28,170 --> 00:01:35,100
The point I'm trying to make is that the doctor will look at various stats to make their diagnosis.
23

24
00:01:35,160 --> 00:01:36,480
They will look at these stats,
24

25
00:01:36,480 --> 00:01:39,960
they will check if these stats are too high or too low,
25

26
00:01:39,990 --> 00:01:44,490
if there's a problem, all in light of your overall condition.
26

27
00:01:44,550 --> 00:01:52,020
Now it turns out that, just like your body, a regression model also has a number of stats that we can
27

28
00:01:52,020 --> 00:01:53,260
look at.
28

29
00:01:53,310 --> 00:01:56,640
In fact there are loads and loads of statistics that we can look at,
29

30
00:01:56,640 --> 00:01:58,210
like a scary amount.
30

31
00:01:58,380 --> 00:02:05,100
But my goal is to show you a couple of these statistics that are relevant to our regression model and
31

32
00:02:05,130 --> 00:02:08,630
also show you how to interpret them and how to make sense of them.
32

33
00:02:08,640 --> 00:02:14,340
The first one we're already quite familiar with because we've looked at r-squared extensively, but as
33

34
00:02:14,340 --> 00:02:19,590
part of this evaluation process I want to introduce you to several other stats as well.
34

35
00:02:19,590 --> 00:02:27,540
For example the p-values of the coefficients, the variance inflation factor and the Bayesian information
35

36
00:02:27,540 --> 00:02:29,010
criterion.
36

37
00:02:29,010 --> 00:02:35,100
We're going to dip our toes into the murky waters of statistics and get a taster for how to evaluate
37

38
00:02:35,400 --> 00:02:36,570
a regression model.