R help - May 2005 - Can simulation involving random number generation be segmented?

Dear all,
Apologies for this pedantic question that only arise when there is hardware
limitation.
Setting: R 2.1.0 for windows xp sp2.
Scenario:
To generate 1000 samples using rnorm for a simulation activity.
Background:
The simulation activity requires so much memory resources that generating
200 samples
clogs up the PF usage as indicated in the Windows Task Manager.
Therefore, short of implementing the simulation on a computer with more
resources,
the alternative is to generate the 1000 samples in 5 separate runs,
each generating 200 samples, closing the R window and re-opening between
runs.
Question to be addressed:
To maintain consistency and ensure reproducibility of the simulation
results, the 1000 samples
generated in one single run should be indentical to the 5x200 samples
generated on 5 separate
runs.
While such consistency can be ensured using set.seed()  in the case of one
single run, in the case
where 5 separate runs are performed, can we do the following to ensure
identical samples being
generated?
1. In the first run, specify the seed by, say, set.seed(1)

2. At the end of the first run, store the .Random.seed by the following
manner:
saved.seed.1<-.Random.seed

3. At the beginning of the second run, assign the saved.seed.1 to
.Random.seed as follows:
.Random.seed<-saved.seed.1

4. At the end of the first run, store the new .Random.seed by the following
manner:
saved.seed.2<-.Random.seed

5. At the beginning of the second run, assign the saved.seed.2 to
.Random.seed as follows:
.Random.seed<-saved.seed.2

This is repeated until 5 runs are completed.

Will the paths of random number generation be identical in these two
approaches? If not, is there
a way to ensure this?

Apologies again for this long-winded inquiry.

Thank you.
Best
Lin

Dr L. Y Hin wrote:
> Dear all,
> Apologies for this pedantic question that only arise when there is hardware
> limitation.
> Setting: R 2.1.0 for windows xp sp2.
> Scenario:
> To generate 1000 samples using rnorm for a simulation activity.
> Background:
> The simulation activity requires so much memory resources that generating
> 200 samples
> clogs up the PF usage as indicated in the Windows Task Manager.
> Therefore, short of implementing the simulation on a computer with more
> resources,
> the alternative is to generate the 1000 samples in 5 separate runs,
> each generating 200 samples, closing the R window and re-opening between
> runs.
> Question to be addressed:
> To maintain consistency and ensure reproducibility of the simulation
> results, the 1000 samples
> generated in one single run should be indentical to the 5x200 samples
> generated on 5 separate
> runs.
> While such consistency can be ensured using set.seed()  in the case of one
> single run, in the case
> where 5 separate runs are performed, can we do the following to ensure
> identical samples being
> generated?
> 1. In the first run, specify the seed by, say, set.seed(1)
> 
> 2. At the end of the first run, store the .Random.seed by the following
> manner:
> saved.seed.1<-.Random.seed
> 
> 3. At the beginning of the second run, assign the saved.seed.1 to
> .Random.seed as follows:
> .Random.seed<-saved.seed.1
> 
> 4. At the end of the first run, store the new .Random.seed by the following
> manner:
> saved.seed.2<-.Random.seed
> 
> 5. At the beginning of the second run, assign the saved.seed.2 to
> .Random.seed as follows:
> .Random.seed<-saved.seed.2
> 
> This is repeated until 5 runs are completed.
> 
> Will the paths of random number generation be identical in these two
> approaches? 
Yes.

Uwe Ligges

If not, is there
> a way to ensure this?
> 
> Apologies again for this long-winded inquiry.
> 
> Thank you.
> Best
> Lin
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html

You might consider the rlecuyer package, which provides parallel
streams and avoids the potential worst case scenario from arbitrary
seed setting.

On 5/25/05, Dr L. Y Hin <lyhin at netvigator.com>
wrote:
> Dear all,
> Apologies for this pedantic question that only arise when there is hardware
> limitation.
> Setting: R 2.1.0 for windows xp sp2.
> Scenario:
> To generate 1000 samples using rnorm for a simulation activity.
> Background:
> The simulation activity requires so much memory resources that generating
> 200 samples
> clogs up the PF usage as indicated in the Windows Task Manager.
> Therefore, short of implementing the simulation on a computer with more
> resources,
> the alternative is to generate the 1000 samples in 5 separate runs,
> each generating 200 samples, closing the R window and re-opening between
> runs.
> Question to be addressed:
> To maintain consistency and ensure reproducibility of the simulation
> results, the 1000 samples
> generated in one single run should be indentical to the 5x200 samples
> generated on 5 separate
> runs.
> While such consistency can be ensured using set.seed()  in the case of one
> single run, in the case
> where 5 separate runs are performed, can we do the following to ensure
> identical samples being
> generated?
> 1. In the first run, specify the seed by, say, set.seed(1)
> 
> 2. At the end of the first run, store the .Random.seed by the following
> manner:
> saved.seed.1<-.Random.seed
> 
> 3. At the beginning of the second run, assign the saved.seed.1 to
> .Random.seed as follows:
> .Random.seed<-saved.seed.1
> 
> 4. At the end of the first run, store the new .Random.seed by the following
> manner:
> saved.seed.2<-.Random.seed
> 
> 5. At the beginning of the second run, assign the saved.seed.2 to
> .Random.seed as follows:
> .Random.seed<-saved.seed.2
> 
> This is repeated until 5 runs are completed.
> 
> Will the paths of random number generation be identical in these two
> approaches? If not, is there
> a way to ensure this?
> 
> Apologies again for this long-winded inquiry.
> 
> Thank you.
> Best
> Lin
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
> 

-- 
best,
-tony

"Commit early,commit often, and commit in a repository from which we can
easily
roll-back your mistakes" (AJR, 4Jan05).

A.J. Rossini
blindglobe at gmail.com