undefined | Better HN

0 pointsorlp3y ago0 comments

It's a neat trick, but can not generate all floating point values. See my other comment for that.

0 comments

The source of disagreement is: his code generates floating point approximations of uniformly-distributed real values in [0,1]; not random floats greater than 0 and smaller than 1.

This distinction has little to do with the subtlety of floats. You can e.g. generate numbers up to 100 with

- 5d20 (5 dice of 20 sides),

- 20d5 or even

- 16d6+1d4.

[Edit: the point of this exercise: these are different representations of the numbers up to 100.]

But even assuming each die with D sides has equiprobable results with prob 1/D, these choices don't have the same probability distribution. To convince yourself of this, compare 1d12 (the probability of getting 1 is 1/12) with 2d6 (the probability of getting 1 is 0).

torotonnato3y ago

If you want max entropy and normally distributed values then, I think, there's no other way to go. Nice post btw

orlpOP3y ago

What do you mean by "normally distributed values"? Your reply also doesn't follow a normal distribution by the way. Did you mean uniform? For uniform values there is a way to go and my other reply exactly describes how.

torotonnato3y ago

Of course, it's uniform. I always swap the two names randomly hehe. I understand your code. The difference with the snippet I posted (which, btw, is not mine) is that the difference between successive floats is always a constant, i.e. there is no rounding. It's fast, has 23 bits of entropy and it's suited for constrained platforms/small demos [0], since it can be coded with very few asm instructions.

I'm not saying that your solution hasn't its merits, well done.

[0] The author of the code is a known member of the demoscene

1 more reply

j / k navigate · click thread line to collapse

0 comments

prionassembly3y ago

The source of disagreement is: his code generates floating point approximations of uniformly-distributed real values in [0,1]; not random floats greater than 0 and smaller than 1.

This distinction has little to do with the subtlety of floats. You can e.g. generate numbers up to 100 with

- 5d20 (5 dice of 20 sides),

- 20d5 or even

- 16d6+1d4.

[Edit: the point of this exercise: these are different representations of the numbers up to 100.]

torotonnato3y ago

If you want max entropy and normally distributed values then, I think, there's no other way to go. Nice post btw

orlpOP3y ago

torotonnato3y ago

I'm not saying that your solution hasn't its merits, well done.

[0] The author of the code is a known member of the demoscene

1 more reply

j / k navigate · click thread line to collapse