CODEWITHSUNDEEP
rrandom-seed
Mark Miller
Mark Miller

Reputation: 13103

Results change depending on whether set.seed is specified before or after loading RData file containing .Random.seed

I am running an R script that reads an external RData file. I am attempting to modify that R script to run without using that RData file to improve generality.

I discovered that the results returned by the script differ depending on whether I use a set.seed statement immediately before or immediately after loading the RData file. I also discovered that the RData file contains a vector called .Random.seed. Otherwise the RData file only contains data as far as I can tell, not any functions or code.

The mere presence of the .Random.seed vector in the RData file appears to change the results of random number generators (runif, rnorm, etc.) or sample statements in the script that appear after I load the RData file even though I never refer to this .Random.seed vector in the script.

How can I reproduce the results of the script that contains the RData file without actually including the RData file in the script?

load("CPT CAL.RData")
.Random.seed[1:5]
#[1]         403         493 -1955963110 -1150619109   212268921
sample(1:100, 10, replace = TRUE)
#[1] 100  60  76  39   7  57  28  85  89  73

sample(1:100, 10, replace = TRUE)
#[1] 81 89 15  4 60 32 58 71 62 17

sample(1:100, 10, replace = TRUE)
#[1] 18 46 89 66 23 24 24 41 37 70

##################################

# Here I try to reproduce the above results using the first element
# of the .Random.seed vector as a seed but this does not work
set.seed(403)
sample(1:100, 10, replace = TRUE)
#[1] 44 31 23 83 16 27 71 64 13 18

sample(1:100, 10, replace = TRUE)
#[1] 86 54 46 25 65 42 76 49 46 50

sample(1:100, 10, replace = TRUE)
#[1] 44 44 90 39 46 14 41 52 36 63

Upvotes: 0

Views: 593

Answers (1)

Mark Miller
Mark Miller

Reputation: 13103

@MrFlick posted a very clear and very helpful explanation for why my original attempt at a solution did not work.

Subsequently I arrived at a solution adequate for my purposes. I placed a set.seed statement immediately after loading the RData file into the original R script. I saved the results as my target values.

Then I added the same set.seed statement into my revised version of the R script that never loads the RData file.

Both approaches returned the same results.

Upvotes: 1

Related Questions