Random Number Generators, .Random.seed and all that..

Martin Maechler Martin Maechler <maechler@stat.math.ethz.ch>
Tue, 1 Dec 1998 09:05:25 +0100

As some of you know,
we have been thinking of allowing the possibility of a
CHOICE of the kind of random number generator (=: RNG) to use in R.
The current R-release snapshot even has some code in it; 
however, this will be changed quite a bit.  Here is a kind of informal
RFC (request for comments / request for criticism / ..):

  1a.  With the new scheme, we still want that
       save(..)  & load(..)  restore the  state of the RNG.

      The new state is  [ RNGkind ; "seed" of that RNG ].
      "seed" really means the state of the chosen RNG

  1b. For compatibility reasons, it makes sense
      to keep this RNGstate in the integer vector  .Random.seed.

      Before each RN generation, .Random.seed is read into the internal
      RNG state variables; afterwards .Random.seed is restored from there.

  1c. In the future,  .Random.seed[1]    == <code of RNGkind>
		      .Random.seed[2:ns] == "seed" itself.

	===> INCOMPATIBILITY to currently saved or set  .Random.seeds.

	We could fudge backward compatibility by

		if(length(.Random.seed) == "desired length for RNGkind" - 1) {

			behave as if .Random.seed[0] == <current RNGkind code>
			.Random.seed[1:n] == <corrected .Random.seed>[2:(n+1)]
	Should I provide this hack?
	If I don't people will more quickly have to change their code,
	setting proper  .Random.seed  whereever they assign one.

  1d. Each RNG has it's own "seed" (state vector).
      Randomize() is only called when .Random.seed isn't there, or
      (new)  if( length(.Random.seed) == 1  && 
		 .Random.seed[1] == "valid RNGkind code" )

  2. Default RNGkind == as now, "Wichmann-Hill"

  3. In addition to .Random.seed, the RNGkind can be queried and set by
     or, e.g.,

	{{ my current code now provides all the above, with the exception of
	   the full backward compatibility hack}}

Comments (incl. praises :-}) welcome!

Martin Maechler <maechler@stat.math.ethz.ch>	http://stat.ethz.ch/~maechler/
Seminar fuer Statistik, ETH-Zentrum SOL G1;	Sonneggstr.33
ETH (Federal Inst. Technology)	8092 Zurich	SWITZERLAND
phone: x-41-1-632-3408		fax: ...-1086			<><

r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch