1
00:00:00,310 --> 00:00:02,250
So then maybe what we can do is just to

2
00:00:02,250 --> 00:00:05,720
pick our test inputs randomly so to do what is called

3
00:00:05,720 --> 00:00:08,850
random testing. And what that means is that we pick the

4
00:00:08,850 --> 00:00:11,720
inputs to test just as we pick a number by rolling

5
00:00:11,720 --> 00:00:15,410
a set of dice randomly. And this will have several advantages.

6
00:00:15,410 --> 00:00:18,780
First, we will pick inputs uniformly. So if we use a

7
00:00:18,780 --> 00:00:21,790
uniform distribution as the basis for our random testing, we will

8
00:00:21,790 --> 00:00:25,540
make no preferences. In other words, all inputs will be considered

9
00:00:25,540 --> 00:00:28,324
equal, of equal value. And what that means in turn, is

10
00:00:28,324 --> 00:00:32,640
that random testing eliminates designer bias. So what does designer bias

11
00:00:32,640 --> 00:00:36,030
mean? Designer bias is the problem of making the same assumption,

12
00:00:36,030 --> 00:00:38,570
when we read the specification and we interpret it and when we

13
00:00:38,570 --> 00:00:42,100
develop test cases. Which means that the developer might develop code,

14
00:00:42,100 --> 00:00:44,930
assuming a given behavior of the user. And we may write

15
00:00:44,930 --> 00:00:47,520
tests, making the same assumptions. And the problem, of course, is

16
00:00:47,520 --> 00:00:50,690
even worse if it's the same person that develops the code and

17
00:00:50,690 --> 00:00:53,730
writes the test cases. With random testing, the problem is gone,

18
00:00:53,730 --> 00:00:57,440
because we just pick randomly what our inputs will be. So

19
00:00:57,440 --> 00:01:00,180
why not do in random? The problem is that when testing,

20
00:01:00,180 --> 00:01:03,610
we are looking for a needle in a haystack. Actually, multiple

21
00:01:03,610 --> 00:01:06,620
needles in multiple haystacks, if we want to be precise. So,

22
00:01:06,620 --> 00:01:09,500
random approaches are not necessarily the best way to go about

23
00:01:09,500 --> 00:01:12,430
it, because we might just be looking in all the wrong

24
00:01:12,430 --> 00:01:15,760
places. So let me show you this, using a different representation

25
00:01:15,760 --> 00:01:18,000
for the haystack. What I'm showing here is a grid, and

26
00:01:18,000 --> 00:01:22,130
imagine this grid just expanding indefinitely outside the screen, and this grid

27
00:01:22,130 --> 00:01:26,120
represents the domain for the program, so each box in the grid,

28
00:01:26,120 --> 00:01:29,050
each square in the grid, it's a possible input. So what happens

29
00:01:29,050 --> 00:01:32,670
with bugs is that bugs are very scarce in this grid. Maybe

30
00:01:32,670 --> 00:01:35,070
there is a bug here, so that means that there is a

31
00:01:35,070 --> 00:01:38,090
bug, than an input, in this point we'll reveal. And maybe there

32
00:01:38,090 --> 00:01:40,820
is another bug that will be triggered by an input over here.

33
00:01:40,820 --> 00:01:44,570
So imagine this spread out over this infinite grid. Its very unlikely

34
00:01:44,570 --> 00:01:47,440
that just by picking randomly that we will be able to get to

35
00:01:47,440 --> 00:01:50,910
these two points. Fortunately not all is lost, there is a silver lining.

36
00:01:50,910 --> 00:01:53,410
So we need to look a little more in depth into this grid.