I'm getting slightly different random numbers depending on the OS (Windows vs Linux) although I have specified the seed using set.seed.
Is there anyway to guarantee reproducibility across platforms?
More specifically, I am looking for a way to force the R on Linux produce the same random numbers the R on Windows produces.
You could test the following code on both platforms:
set.seed(1234); print(rnorm(1), 22)
Please look at the following links, the answers are given:
1- https://mathematica.stackexchange.com/questions/157621/does-seedrandom-give-the-same-set-of-random-numbers-across-operating-systems
2- ttps://softwareengineering.stackexchange.com/questions/204850/can-a-random-number-generator-ever-produce-different-output-given-identical-seed
3- https://stackoverflow.com/questions/40361041/achieve-same-random-number-sequence-on-different-os-with-same-seed
====
Good luck
Thank you for the reply. The R version is the same on both platforms.
Here is the output of running "version" command on both platforms.
Linux
-----------
platform x86_64-pc-linux-gnu
arch x86_64
os linux-gnu
system x86_64, linux-gnu
status
major 3
minor 4.3
year 2017
month 11
day 30
svn rev 73796
language R
version.string R version 3.4.3 (2017-11-30)
nickname Kite-Eating Tree
Windows
--------------
platform x86_64-w64-mingw32
arch x86_64
os mingw32
system x86_64, mingw32
status
major 3
minor 4.3
year 2017
month 11
day 30
svn rev 73796
language R
version.string R version 3.4.3 (2017-11-30)
nickname Kite-Eating Tree
You can see one random number generated on the two platforms below:
Linux:
> set.seed(1234); print(rnorm(1), 22)
[1] -1.207065749385421105089
Windows:
> set.seed(1234); print(rnorm(1), 22)
[1] -1.2070657493854209
Did u get such an answer from the beginning ? I mean what was the problem that u didn't manage the get the same answers in the beginning, or in other words, what solved the problem ?
Thank you, Vahid. Do you know how I could do this in R? I could not find the right command for R in order to get the same result from the rnorm command on both platforms. I tested some of the suggestions, but no luck.
Dear Vahid,
My problem was that I simulated a couple of data sets, and I ran some analysis on them. I then decided to use cluster computers to run the same analysis on the same data sets. I noticed I was getting different results. So I broke the code to small parts and checked to see what was causing the different result. I found out that it was because of the different random numbers which were being generated.
Now, if I am able to produce the same result for a command like rnorm on both platforms, my problem will be solved.
As you can see in the code I posted above, the generated numbers are slightly different. But they are making a huge difference in my analysis.
Did you try adding "set.seed(10)" (or your favorite number) before your analyses in both platform. If both use the same random number generation process, it should make the job. (see ?set.seed for more details)
Sorry, I did not read enough the previous answers... the set.seed command was already there. Oups!
Hi Farideh
Thanks for the explanations. I haven't tested this in R.
What can be observed from your results is the following:
Linux : -1.207065749385421105089
Windows: -1.2070657493854209
====
Maybe one solution could be to limit the number of digits after decimal point ? As u can see it seems that two platforms are returning up to 14 similar digits after the decimal point. So, perhaps if you use up to that range, the problem will be solved, am I right ?
Hi Farideh
I don't think you can do that even on the same system as the moment you run a program the random number will change because as the name suggest they are random in nature. To be more exact pseudo random.
Sahay
Thank you, Vahid. I considered this solution, but the problem is that I have already run lots of my simulations on my laptop (Windows). As you see in the example above, even if I truncate the numbers produced by Linux, they will not be the same as those produced by Windows. So, I have to re-run all my simulations again using Linux which is not a big deal. But I was hoping for a more general solution. Because you can never be sure if you truncate a number up to the 10th decimal, the 9th decimal would be the same by both operating systems.
Thank you Marius and Shashwat.
Dear Shashwat, setting the seed for random number generation will help us the same random number for the purpose of reproducibility. Is that what you are talking about?
I think Vahid already posted the answer referencing stackoverflow [3].
It is suggested to use a certain implementation such as mt19937.
Apparently R has a package which interfaces the dieharder suite (http://webhome.phy.duke.edu/~rgb/General/dieharder.php):
https://cran.r-project.org/web/packages/RDieHarder
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R