spdegabrielle
2021-2-11 12:01:10

Did it 4 times, and each declaration (with captcha) seemed to be about declaring the site (mirror.) was safe, not the download specifically.


spdegabrielle
2021-2-11 12:11:40

Its quite confusing to accept - I could do a short video for youtube, showing how to get through the download and install. (I have camtasia on new work PC)


gknauth
2021-2-11 14:17:53

Downloaded at work in just a few seconds.


gknauth
2021-2-11 14:21:39

gknauth
2021-2-11 14:21:54

gknauth
2021-2-11 14:21:59

gknauth
2021-2-11 14:22:08

gknauth
2021-2-11 14:27:18

Win10


spdegabrielle
2021-2-11 14:28:19

register with symantec?


mathieu
2021-2-11 15:23:53

@mathieu has joined the channel


anything
2021-2-11 15:24:55

That could be a typo? As random’s documentation says, Racket uses Pierre l’Ecuyer’s https://pubsonline.informs.org/doi/pdf/10.1287/opre.47.1.159\|MRG32k3a. It seems that under a certain set of conditions (specified in the paper), the precision would be 53 bits, not 54. FWIW, Python’s default PRNG, which is Mersenne Twister, also seems to provide 53 bits of precision in machines with IEEE 754. Maybe someone here with more knowledge can give you a final word on this.


mflatt
2021-2-11 17:26:55

I think the “54” in the docs is not helpful or useful. Internally, the implementation needs 54 bits (53 plus sign) to fit in a double , but the state space is more like 256 bits, because the state is close to 32 bits in each of 8 doubles.


mflatt
2021-2-11 17:27:34

I must not have that quite right — there are only 6 doubles. But somehow I thought it worked out to 256-ish bits


mflatt
2021-2-11 17:36:26

256 must be wrong, The state is 6 numbers that fit in 32 bits that don’t use the full range, so it’s under 6 * 32 = 192 and more than 6 * 31 = 186.


badkins
2021-2-11 17:41:26

Thanks @mflatt, that should be plenty for my purposes (Zobrist hash for chess engine). I think 64 bits reduces the collisions to an acceptable level, so (random) should work fine.