I really haven't looked enough at Roger's material on his random number
generator but the following seems to me to be necessary.
The output sequence depends on the inital state. Now it becomes important
to start some large number of steps down that sequence, where the
'distance' is easily set. This allows you to return to the same position
in the sequence for testing, but also allows different users to have an
easily controlled different starting point.
When I read of a starting seed I wonder if I have only two choices, start
with that seed or run some artifical number of hits on the generator. I
suppose you can set the seed 'randomly' but that doesn't change the other
'hidden' numbers behind the generator.
I don't have a good answer, and one of Roger's comments did address a
generator like this of mine (there were better ones around). There are four
inherent loops in the generator and you can specify the number of times
each loop gets hit, and still have an easy way to return to your starting
point. These loops control the internal machinations so that different
users start differently. An earlier article about generators pointed out
that it important to have members of a group using the same generator but
different sequences.
I am reminded of this only by recent comments about '?' or '?.'
Ralph Selfridge
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
