jbs@WATSON.IBM.COM (05/13/91)
Herman Rubin writes:
I believe most people are aware of the existence of simulation, including
Monte Carlo, or Las Vegas, methods for obtaining answers to otherwise
intractable problems. In these, it is almost always the case that
considerable use must be made of non-uniform random variables; the
quantities of interest are usually exponential, normal, Cauchy, binomial,
Poisson, etc., and not uniform.
In any particular problem, it is quite likely that thousands or millions or
even billions of these random numbers will be used. If this is not a situation
which calls for efficiency, it will do until a real one comes along. :-)
Even granting that Monte Carlo methods are important I remain
unconvinced that:
1) The time spent generating random numbers dominates these
computations.
2) Your proposed instructions would reduce the time it takes
to generate the random numbers.
James B. Shearerhrubin@pop.stat.purdue.edu (Herman Rubin) (05/13/91)
In article <9105130247.AA09317@ucbvax.Berkeley.EDU>, jbs@WATSON.IBM.COM writes: .................... > Even granting that Monte Carlo methods are important I remain > unconvinced that: > 1) The time spent generating random numbers dominates these > computations. Sometimes it does, and sometimes it doesn't. > 2) Your proposed instructions would reduce the time it takes > to generate the random numbers. The algorithms I envision would definitely win, even with the present architecture, if enough accuracy was wanted. But even a rather simple algorithm, using on the average less than a dozen bits, is likely to crawl on the present architecture. -- Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399 Phone: (317)494-6054 hrubin@l.cc.purdue.edu (Internet, bitnet) {purdue,pur-ee}!l.cc!hrubin(UUCP)