How many times to use set.seed(123)

Question

I have 60 lines of code. Throughout the code there are several calls to random number generators including rnorm(). Is it enough to put set.seed(x) in the very beginning of the code or do I need to set.seed every time random number generation occurs in the code?

Answer 1

It really depends on how you foresee the code changing in the future.

If you expect that you will be including commands at an earlier point in the code that will require random number generation and you want to replicate the results you were getting earlier before inserting that code, you should use set.seed() at the appropriate points in your code.

Example:

set.seed(1)
A <- rnorm(10)
B <- rnorm(10)
C <- rnorm(10) ## I always want "C" to be the results I get here

set.seed(1)
AA <- rnorm(10); BB <- rnorm(10); CC <- rnorm(10)

identical(A, AA)
# [1] TRUE
identical(B, BB)
# [1] TRUE
identical(C, CC)
# [1] TRUE

set.seed(1)
A <- rnorm(10); B <- rnorm(10); C <- rnorm(10)

set.seed(1)
AA <- rnorm(10); BB <- rnorm(10); BA <- rnorm(10); CC <- rnorm(10)

identical(A, AA)
# [1] TRUE
identical(B, BB)
# [1] TRUE
identical(C, CC)
# [1] FALSE

In the above, if I wanted "C" to always be the same no matter what comes before it, I should set the seed just before that.

Note that since I hadn't reset the seed before creating C or CC, and there is a new function requiring random number generation between BB and CC in the second example, the values for C and CC are now different. If you wanted them to be the same, you would have to insert another set.seed just before creating C and CC, as follows:

set.seed(1)
A <- rnorm(10)
B <- rnorm(10)
set.seed(2)
C <- rnorm(10) ## I always want "C" to be the results I get here

set.seed(1)
AA <- rnorm(10); BB <- rnorm(10); BA <- rnorm(10);
set.seed(2)
CC <- rnorm(10)

identical(A, AA)
# [1] TRUE
identical(B, BB)
# [1] TRUE
identical(C, CC)
# [1] TRUE