| Re: Bizarre (?) behavior of randomu [message #57829] |
Fri, 04 January 2008 06:44 |
weitkamp
Messages: 33
Registered: October 1998
|
Member |
|
|
John,
As the two previous posters have already mentioned, it is true that
the heart of the problem is that every time you run "rannummers" anew,
the "seed" value is undefined at the beginning, i.e., when the loop
starts at k=0. Allard de Wit is right in that in this case (a call to
RANDOMU with the seed variable undefined) "the seed value will be
taken from some system values (like the systime or similar)".
It is strange indeed that this leads to the repetition pattern you
describe (btw, I reproduced this behavior on two different Linux
systems with IDL 6.3 and 6.4). But since the behavior you observe
results from an ill-defined initial state, there is probably no point
in reasoning why it turns out just this way and not another.
As the two previous posters have pointed out, the solution is to
carefully select the value of "seed" each time you call RANDOMU,
depeding on what you want:
* To obtain a first pseudo-random number (or number sequence), input
the seed variable with a well-defined long integer scalar.
* "To continue the pseudo-random number sequence, input the variable
containing the saved state [i.e., the value of "seed" given back by
RANDOMU] as the Seed argument in the next call to RANDOMN or
RANDOMU." (quoting the IDL 6.4 online help).
* If you want to obtain an exact repetition of a number sequence
obtained in this way, start the same code again, with the same seed
value passed in the initial call to RANDOMU as before.
I certainly don't think that a common block or a wrapper object is
needed for this -- adding a single positional parameter "seed" to the
procedure does the job:
pro rannummers, seed
nn=100000
kmax=4
for k=0,kmax-1 do begin
rnum=randomu(seed,nn)
print,rnum[0:5],format='(6f10.5)'
endfor
end
IDL> seed = 478319L
IDL> rannummers, seed
% Compiled module: RANNUMMERS.
0.55458 0.71883 0.50224 0.45372 0.04597 0.01121
0.55780 0.51784 0.57281 0.17167 0.53495 0.70408
0.31777 0.72242 0.59234 0.98761 0.35242 0.87670
0.55746 0.99774 0.43512 0.06631 0.96045 0.63092
IDL> rannummers, seed
0.29667 0.92210 0.57792 0.39947 0.91916 0.63813
0.76520 0.73514 0.77973 0.11008 0.32726 0.94418
0.96893 0.69209 0.11459 0.27642 0.73942 0.57509
0.37667 0.64760 0.04895 0.66316 0.75020 0.77216
IDL> rannummers, seed
0.10873 0.68996 0.19374 0.47691 0.84283 0.73352
0.99465 0.27692 0.61346 0.59465 0.13965 0.38272
0.01367 0.72033 0.74071 0.48875 0.66167 0.99068
0.10014 0.64916 0.70723 0.11969 0.66077 0.59047
IDL> seed = 478319L
IDL> rannummers, seed
0.55458 0.71883 0.50224 0.45372 0.04597 0.01121
0.55780 0.51784 0.57281 0.17167 0.53495 0.70408
0.31777 0.72242 0.59234 0.98761 0.35242 0.87670
0.55746 0.99774 0.43512 0.06631 0.96045 0.63092
In the above example, the three first calls produce a true pseudo-
random series of numbers (without any visible repetition scheme). The
last call produces the same series as the first one because we seeded
the number generator with the same value.
Cheers,
Timm
Allard de Wit wrote:
> The trick is indeed to specify a random seed during the first call to
> RANDOMU and preserve the seed value through subsequent calls to
> RANDOMU. If you do not specify it in the first call (like in your
> example) the seed value will be taken from some system values (like
> the systime or similar) and RANDOMU will produce different sets of
> random numbers with each call. This is generally undesirable,
> if you want your results to be reproducible.
>
> example:
> IDL> seed=1
> IDL> print, randomu(seed, 10)
> 0.415999 0.0919649 0.756410 0.529700 0.930436
> 0.383502 0.653919 0.0668422 0.722660 0.671149
> IDL> print, randomu(seed, 10)
> 0.383416 0.631635 0.884707 0.519416 0.651519
> 0.237774 0.262453 0.762198 0.753356 0.909208
> IDL> seed=1
> IDL> print, randomu(seed, 10)
> 0.415999 0.0919649 0.756410 0.529700 0.930436
> 0.383502 0.653919 0.0668422 0.722660 0.671149
> IDL> print, randomu(seed, 10)
> 0.383416 0.631635 0.884707 0.519416 0.651519
> 0.237774 0.262453 0.762198 0.753356 0.909208
>
> Note that while I specify seed=1, it will be replaced by a LONARR(36)
> after the first call to RANDOMU.
>
> A more elegant solution without using COMMON blocks is to wrap the
> random number generator in an IDL object.
>
> with best regards,
>
> Allard
|
|
|
|
| Re: Bizarre (?) behavior of randomu [message #57831 is a reply to message #57829] |
Fri, 04 January 2008 01:54  |
wita
Messages: 43
Registered: January 2005
|
Member |
|
|
The trick is indeed to specify a random seed during the first call to
RANDOMU and preserve the seed value through subsequent calls to
RANDOMU. If you do not specify it in the first call (like in your
example) the seed value will be taken from some system values (like
the systime or similar) and RANDOMU will produce different sets of
random numbers with each call. This is generally undesirable,
if you want your results to be reproducible.
example:
IDL> seed=1
IDL> print, randomu(seed, 10)
0.415999 0.0919649 0.756410 0.529700 0.930436
0.383502 0.653919 0.0668422 0.722660 0.671149
IDL> print, randomu(seed, 10)
0.383416 0.631635 0.884707 0.519416 0.651519
0.237774 0.262453 0.762198 0.753356 0.909208
IDL> seed=1
IDL> print, randomu(seed, 10)
0.415999 0.0919649 0.756410 0.529700 0.930436
0.383502 0.653919 0.0668422 0.722660 0.671149
IDL> print, randomu(seed, 10)
0.383416 0.631635 0.884707 0.519416 0.651519
0.237774 0.262453 0.762198 0.753356 0.909208
Note that while I specify seed=1, it will be replaced by a LONARR(36)
after the first call to RANDOMU.
A more elegant solution without using COMMON blocks is to wrap the
random number generator in an IDL object.
with best regards,
Allard
|
|
|
|
| Re: Bizarre (?) behavior of randomu [message #57846 is a reply to message #57831] |
Wed, 02 January 2008 14:24 |
M. Katz
Messages: 69
Registered: May 2005
|
Member |
|
|
The trick is to preserve the 'seed' from one run to the next.
If you put the seed in a common block, for example, then it won't keep
getting reset every time you run the program.
So the first two lines of your program would become
pro rannummers
common rblock, seed ;--- common blocks need to be named.
I'm not an expert in (pseudo) random number generation, but I believe
that the whole point of the 'seed' argument is for when people need to
re-generate the exact same sequence of random numbers. You can see by
a simple test that if you give randomu() the same seed value, you get
the same sequence of numbers--every time.
Consequently, every time you restart IDL and run this program, you'll
get the same sequence of random numbers. That's the expected behavior.
Since you want different results every time, you have to set the seed
to a different value each time. For that, you could use the system
clock {systime(1)}, or some such thing you choose. Preserve the seed
using the common block, as above, and then you're all set.
Usually common blocks are frowned upon as a programming technique.
It's a sneaky way of passing information around that bypasses
arguments and keywords. Yet in this case, I think it's justifiable.
M.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
