a*****n
发帖数: 5158 | 1 z=a+c+b-c
的error为啥与z=a+b不一样? |
y******7
发帖数: 4 | |
j**p
发帖数: 53 | 3 the first c and the 2nd c may come from different samples, hence even
though they are theoretically equal, numerically the first c might
be c_theory + 0.5 while the 2nd c is c_theory-0.5, so you get a +1 error
here.
【在 a*****n 的大作中提到】
: z=a+c+b-c
: 的error为啥与z=a+b不一样?
|
w********o
发帖数: 10088 | 4 round-off error?
【在 a*****n 的大作中提到】
: z=a+c+b-c
: 的error为啥与z=a+b不一样?
|
s*******n
发帖数: 1474 | 5 ......
先看标题啊
【在 w********o 的大作中提到】
: round-off error?
|
j**p
发帖数: 53 | 6 if you'd like a bit more rigor, here it is :
(let's drop a and b; that wouldn't change the essense of the discussion)
z=c-c is actually c_1 - c_2, where c_1 and c_2 are independent random
variables with identical distribution. So, even though z has zero mean, its
variance is nonzero (Var[z]=Var[c_1]+Var[c_2]=2Var[c]; i.e. the std of z is
sqrt(2)*(std of c)) |
