Somewhat related, but humans are also REALLY bad at generating randomness and one of our big tells is an aversion to repeats. If you ask someone to pick 0-9 randomly, repeatedly, they will rarely repeat numbers. But in a truly random sample a repeat is likely 10% of the time, and a three-peat will happen roughly once every 100 numbers. An average person will rarely if ever repeat a number and sure as heck won't ever come up with it three times in a row.
Quasi-Monte Carlo has a rate of convergence close to O(1/N), whereas the rate for the Monte Carlo method is O(N^(−0.5))
For such applications it's best to use quasi-random numbers (a.k.a. low-discrepancy sequences) such as the Halton sequence or the Sobol sequence instead of pseudorandom numbers.
I think it’s actually the prototypical real-world software engineering problem. User says they want X (random music). X is a term in software, so you give them that (you get a random song). They’re not happy. You dig and find out they really want A, B, and C (next song is unknown, songs don’t repeat too soon or too infrequently). This new problem is harder to verify (how soon is too soon?).
Editing in tips on solving this sort of problem. You can turn vague requirements into precise requirements. Rather than make the precise requirements exactly equivalent to the vague ones, it's easier to make them more restrictive. Is playing a song again within 50% of the length of the playlist "too soon"? Maybe. How about within 80% of the length of the playlist? Definitely not. We can give ourselves the requirement "Songs must always play again between 80% and 125% of the length of the playlist." Much easier to solve, much easier to test.
Sometimes the extra restriction make the problem harder (not usually I've found). Still, this is a great trade because understanding requirements is harder than solving well defined problems.
[To the point of this whole post] Requirements can be turned into testable properties even if it's not programmatic. "When I look at a list of chosen songs, there must be no obvious patterns." Who says what's obvious? You do! Then, have someone else do the same.
Consider extreme cases. Extreme cases tend to be the most or least important. If they're least important, create a new set of easier requirements or drop it all together. "If 3 - 10 songs, always play within double the playlist, no obvious patterns, never twice in a row. If 2 alternate, if 1 repeat."
Took months and a lot of patience to extract that crucial piece of information from the customer. I actually coded a recursive algorithm which to my surprise generated every possible (three letter) sequence in an acceptable sequence, enough for them for several lifetimes.
> Extreme cases tend to be the most or least important.
Yeah, that's my suspicion.
> come close to outputting all possible numbers once before getting into re-runs
In a dice rolling game, you want as close to true random as your PRNG can get. In card drawing, you typically want EXACTLY all possible cards once before getting "repeats". Where do you want something in between?
What you're talking about is "with/without replacement".
Anecdote time:
A few years ago with my friends we were discussing how easy is to roll 5 dices simultaneously and get the same result in all of them. This is a possible way to win a popular game here https://en.wikipedia.org/wiki/Generala We estimated how often you can roll the dices and the probability, and we estimated that you must try during 2 or 3 hours to get that result.
We were young, had a lot of free time, so one of my friend started rolling 5 dices while he was talking with us and eating. After about 2 hour he got the 5 equal dices in a roll. (IIRC he tried again, with a similar result.)
(Note that 2 or 3 hours is consistent with how this outcome is used in the game to win automatically. It's possible to get this in a normal game, but it's not super common.)
Also, each time you add a dice, the time increase exponentially. With 6 dice it's 14-21 hours, like a day. With 7 dice it's like a week. With 9 dice is like a month. With 10 dice is half a year. 1000 dice will take longer (and x6 if you want to get only 6s instead of any repeated number).
(Note that parallelizing this to all humans only adds 14 dices. If everyone on Earth start rolling 24 dices, it will take like half a year until one of us get 24 equal dices.)
https://www.quantamagazine.org/elegant-six-page-proof-reveal...
thanks for nothing.
This intersects with the concept of entropy: assuming that you have a box containing a gas whose particles move randomly about the volume of the box, then at some point you take a snapshot of the position of every single particle in the box and you discover that they're all in the right half of the box, the left side being in a vacuum. Would you assume that it's just random chance? It could be. It certainly isn't.
Meanwhile any of the trillions and trillions of snapshots showing particles more or less uniformly distributed within the box are all more "random looking" and are what is expected from such an experiment. These configurations as a group occupy the vast majority of the phase space for the contents of the box.
(Looks like my profile is out of date, by the way.)
The sequence length of a single die roll is 1. The probability of any particular roll is the same.
Let's do the next part with two-sided dice because it will be quicker.
The sequence length of a roll of two dice is 2. The roll is a bag; the probability of either [0] or [1] is lower than the probability of [1,2] because it's a set membership and [1,2] has the same members and is the same as [2,1]
If the sequence is long enough you can model how likely it is to have been produced by a fair die. Are all numbers equally distributed? Are some numbers more likely to follow or not follow other numbers? Are some patterns repeating?
Of course any sequence can be produced by a fair die, but you can still create some objective metric that will tell you how truly random a sequence is, and the longer it is the more accurate it will be. It's what tests like Diehard are about after all.
Can you roll a fair die a thousand times and only get 6s? Well yes of course. It'll never happen though.