topic Re: How to create a random number seed at the start of a simulation in SAS Programming

How to create a random number seed at the start of a simulation

PaigeMiller — Fri, 04 Oct 2013 17:56:42 GMT

I have a large simulation, that involves several data steps and a few loops.

I would like to set a single random number seed at the start of the simulation, and thus have a single reproducible stream of random numbers. How can I do this?

Here is the simulation, in pseudo-code

Loop 1

Data step A creating uniform random variables

Loop 2

Data step B creating more uniform random variables

More PROCs and Data steps

Data step C creating more uniform random variables

PROC SOMETHING

End of Loop 2

End of Loop 1

Re: How to create a random number seed at the start of a simulation

data_null__ — Fri, 04 Oct 2013 18:16:33 GMT

Perhaps you want to save the last SEED used at each step/interation as the SEED for the next.

If you use the CALL version of the random number function you can track the seed.

Also SURVEYSELECT will create macro variables with seed info.

proc surveyselect data=sashelp.class out=sample rate=.5; 
   run; 
%put NOTE: &=SYSRANDOM &=SYSRANEND;

Re: How to create a random number seed at the start of a simulation

PGStats — Fri, 04 Oct 2013 18:53:33 GMT

In fact, all procedures (or almost) will update the SYSRANEND macro variable. Although not perfect (check the documentation) this seems to be the best method to carry a seed from one procedure to the next. - PG

Re: How to create a random number seed at the start of a simulation

art297 — Fri, 04 Oct 2013 18:55:26 GMT

Paige,

I had come up with the same idea as DataNull did .. i.e., capturing the latest seed in a datastep using call ranuni. The following code shows that it produces the same result as I think you are trying to achieve:

data part1;

seed=872643;

do i=1 to 10;

call ranuni(seed,random);

output;

end;

call symput('seed',seed);

run;

data part2;

seed=&seed.;

do i=1 to 10;

call ranuni(seed,random);

output;

end;

call symput('seed',seed);

run;

data part3;

seed=&seed.;

do i=1 to 10;

call ranuni(seed,random);

output;

end;

call symput('seed',seed);

run;

data test;

do i=1 to 30;

random=ranuni(872643);

output;

end;

run;

HTH,

Art

Re: How to create a random number seed at the start of a simulation

data_null__ — Fri, 04 Oct 2013 18:57:05 GMT

Example of using last seed as next seed with data steps.

%let seed=121853;
data one;
   set sashelp.class end=eof;
   retain seed &seed;
   if _n_ eq 1 then put 'NOTE: ' seed=;
   call ranuni(seed,y);
   run; 
proc print; 
   run; 
data two;
   if _n_ eq 1 then do; 
      set one(keep=seed) point=nobs nobs=nobs;
      put 'NOTE: ' seed=;
      end; 
   set sashelp.class;
   call ranuni(seed,y2);
   run; 
proc print; 
   run; 

Re: How to create a random number seed at the start of a simulation

PaigeMiller — Fri, 04 Oct 2013 19:01:34 GMT

Thanks to all, looks like I have several solutions.

I think I just need a hotline to , and , and should just skip the forum entirely

Re: How to create a random number seed at the start of a simulation

Linlin — Fri, 04 Oct 2013 19:21:30 GMT

Paige,

You got great solutions in an hour. You are the luckiest man today and should go out to buy lottery tickets.

Re: How to create a random number seed at the start of a simulation

data_null__ — Fri, 04 Oct 2013 19:22:58 GMT

Using SYSRANDOM and SYSRANEND Macro Variables to Produce Random Number Streams

Re: How to create a random number seed at the start of a simulation

PaigeMiller — Fri, 04 Oct 2013 19:23:50 GMT

Linlin wrote:

Paige,

You got great solutions in an hour. You are the luckiest man today and should go out to buy lottery tickets.

Statisticians don't believe in purchasing lottery tickets, the expected gain is negative, worse than just putting money in a savings account that pays 0.5% annually.

Re: How to create a random number seed at the start of a simulation

art297 — Fri, 04 Oct 2013 20:19:41 GMT

, Sure am glad I'm not a statistician. Conversely, putting money into a savings account that pays 0.5% annually, only has the potential of resulting in a loss when one considers the effects of inflation and the taxes one has to be on the interest. While the likelihood of winning the grand prize of tomorrow's Canadian lottery ($26 million) has rather poor odds ( 1 in 13,983,816) the potential IS a bit larger than a savings account, the psychological excitement about the possibility of winning exists and, in the unlikely event that one does actually win, the winnings are tax free.

Re: How to create a random number seed at the start of a simulation

PaigeMiller — Fri, 04 Oct 2013 20:29:33 GMT

Compared to my friends who regularly play the lottery and football pools and the like, I have not lost a single penny, while they have lost thousands of dollars. I like my situation better than theirs.

My ex-mother-in-law bought 5 dollars in Lottery tickets every week that I knew her. Once I asked her if she ever won anything, and she very enthusiastically replied that Yes, once, she won $50.

Good luck with your lottery tickets.

Re: How to create a random number seed at the start of a simulation

ballardw — Fri, 04 Oct 2013 22:51:35 GMT

I'm wondering where I can get 0.5% on a savings account. Locally with my balance it's closer to 0.01%.

Of course if I only buy a lottery ticket with the max payout is large enough then the expected winning become positive. But living long enough to play enough times ...:smileylaugh:

Re: How to create a random number seed at the start of a simulation

PGStats — Sat, 05 Oct 2013 17:25:34 GMT

One quote I like:

"Limits on knowledge and computation are central to the human condition."

by Herbert A. Simon in Models of Bounded Rationality: Empirically grounded economic reason, MIT Press, 1982

Many profitable industries depend on those limits: insurance, lotteries and casinos, marketing, stock trading, to name a few. And I never forget that some might want to add statistics to the list.

Re: How to create a random number seed at the start of a simulation

Rick_SAS — Wed, 09 Oct 2013 18:34:42 GMT

This is covered on p. 33-34 of my book. Assuming that "Loop 1" is a macro loop, I think you can get rid of it and gain some efficiency in your simulation by using BY processing. See Chapter 6 of my book or see the article Simulation in SAS