There are several ways to generate random numbers in R package “parallel”. Here I used mcmapply as an example to show how to generate 5 random numbers from 4 cores.
Same seed for each core
When you set MC.set.seed as FALSE, each core has the same seed and generate the same random numbers.
set.seed(123)
mcmapply(rnorm, n=rep(5,4), mc.set.seed = FALSE, mc.cores = 4)
As shown below, 4 cores generate 4 series (4 columns) of random numbers. The 4 columns are the same because they have same seed.
## [,1] [,2] [,3] [,4]
## [1,] -0.9685927 -0.9685927 -0.9685927 -0.9685927
## [2,] 0.7061091 0.7061091 0.7061091 0.7061091
## [3,] 1.4890213 1.4890213 1.4890213 1.4890213
## [4,] -1.8150926 -1.8150926 -1.8150926 -1.8150926
## [5,] 0.3304096 0.3304096 0.3304096 0.3304096
Different seeds for each core
When you set MC.set.seed as TRUE, each core has different seeds and will generate different random numbers.
Without reproducibility
There are several methods in R to generate random numbers, the default method is “Mersenne-Twister”. With this method, we cannot get the same random numbers with the same seed.
RNGkind("Mersenne-Twister")
set.seed(123)
mcmapply(rnorm, n=rep(5,4), mc.set.seed = TRUE, mc.cores = 4)
For example, we run the code above and get the following random numbers.
## [,1] [,2] [,3] [,4]
## [1,] -0.2261798 -0.6251189 -2.37868583 1.20695891
## [2,] 2.4251365 0.3667103 1.54679159 -0.09272884
## [3,] 0.7750225 -1.5508046 -0.56366992 -0.46451586
## [4,] 0.0124790 -1.5446089 -0.62513942 -0.73186601
## [5,] 0.8173900 -1.2199431 0.07999375 0.96084955
The columns are different because each core has different seeds. If we run the same code again, we will get different random numbers even we still set the same seed:
## [,1] [,2] [,3] [,4]
## [1,] 2.2683726 0.1390017 -1.0274918 -0.28896864
## [2,] -0.4140674 2.1043673 -2.7126309 -0.62731623
## [3,] -0.4327392 0.8579146 0.8667401 0.24681518
## [4,] 0.3623787 0.9685905 0.2008967 0.04581639
## [5,] 1.3556560 0.4316045 0.3484889 -0.16804107
With reproducibility
To solve the reproducibility problem, we need to set the random number generator method to “L’Ecuyer-CMRG” and use the following code.
RNGkind("L'Ecuyer-CMRG")
set.seed(123)
mc.reset.stream()
mcmapply(rnorm, n=rep(5,4), mc.set.seed = TRUE, mc.cores = 4)
Run for the fisrt time:
## [,1] [,2] [,3] [,4]
## [1,] -0.4094454 -0.48906078 -1.0388664 0.7613014
## [2,] 0.8909694 0.43304237 1.5745125 2.2994158
## [3,] -0.8653704 -0.03195349 0.7470820 0.2002062
## [4,] 1.4642711 0.14670372 0.6718720 -0.2975786
## [5,] 1.2674845 -1.75239095 0.2691436 0.4608767
Run for the second time:
## [,1] [,2] [,3] [,4]
## [1,] -0.4094454 -0.48906078 -1.0388664 0.7613014
## [2,] 0.8909694 0.43304237 1.5745125 2.2994158
## [3,] -0.8653704 -0.03195349 0.7470820 0.2002062
## [4,] 1.4642711 0.14670372 0.6718720 -0.2975786
## [5,] 1.2674845 -1.75239095 0.2691436 0.4608767
The same!
