Contrived seeds in `xoshiro256**`

60 views
Skip to first unread message

qscgukax...@gmail.com

unread,
Feb 2, 2020, 1:00:42 PM2/2/20
to prng

WHAT?

This is mind blowing.

What xoshiro variation should I use to fix this? Will xoshiro512++ do?

Sebastiano Vigna

unread,
Feb 2, 2020, 4:47:05 PM2/2/20
to prng


> On 2 Feb 2020, at 19:00, qscgukax...@gmail.com wrote:
>
> http://www.pcg-random.org/downloads/snippets/corrxoshiro.c
>
> WHAT?
>
> This is mind blowing.

There's really no limit to the shit that Melissa O'Neill can generate (and to the number of people that fall for it!) 😂😂😂😂😂😂.

Try replacing the main() in that program with the main() below.

Guess what? The second initial state is a few million steps from the first initial state. So you're basically printing twice the same sequence. It is not so surprising that the resulting mix fails statistical tests...

Ciao,

seba


PS: In case you have been fooled by Melissa O'Neill, know that when one speaks of "independence of subsequences" there is an implied "nonoverlapping", or the requirement makes no sense. Melissa O'Neill has proven in the past to be very smart at playing with people expectations, removing silently implied assumptions (e.g., that the state of a generator must have good mixing in her crazy "ext" generators).



int main(int argc, char* argv[]) {
// Seed data fresh from /dev/random
uint64_t xo1[4] = { 0x100eeaf0461d14bb, 0xc67ba3a655084ee9,
0xcd64d86f417ed16f, 0x125247c9fa0e6ed0 };

for(int i = 0; i < 3806501; i++) next(xo1);

// More random-looking seed data
uint64_t xo2[4] = { 0xe874f9b004b1beaf, 0x7ddf99501670f4ab,
0x4db1ee8a7057915e, 0xe11fec18143a962f };

for(int i = 0; i < 4; i++) printf("0x%llx ", xo1[i]);
printf("\n");
for(int i = 0; i < 4; i++) printf("0x%llx ", xo2[i]);
printf("\n");

}

Yhgrrr fhyr45t

unread,
Feb 3, 2020, 2:11:16 AM2/3/20
to prng
That's like correlating a true random number generator with two universes a few seconds apart.

Yhgrrr fhyr45t

unread,
Feb 3, 2020, 2:13:55 AM2/3/20
to prng

Yhgrrr fhyr45t

unread,
Feb 3, 2020, 2:19:11 AM2/3/20
to prng
Also, the table at http://www.pcg-random.org looks exactly like some sort of antivirus comparison.
Reply all
Reply to author
Forward
0 new messages