Timeline for Sandbox for Proposed Challenges
Current License: CC BY-SA 4.0
29 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 9, 2020 at 17:03 | history | edited | Wheat WizardMod | CC BY-SA 4.0 | deleted 820 characters in body |
| Feb 25, 2020 at 20:55 | comment | added | FryAmTheEggman | You could consider "Love all integers equally" or "All numbers are interesting" or something for sillier challenge names. I think this works pretty well as-is. However, because of the optional input, I would recommend reiterating that the input could also be assumed to be empty (as a variant of no input). A few languages can get hung up on that (though I can't think of one I'd like to use for this). | |
| Feb 25, 2020 at 20:47 | history | edited | FryAmTheEggman | CC BY-SA 4.0 | Fixed assumed typo |
| Feb 25, 2020 at 14:01 | history | edited | Wheat WizardMod | CC BY-SA 4.0 | added 270 characters in body |
| Feb 25, 2020 at 13:34 | comment | added | Jonathan Frech | Let us continue this discussion in chat. | |
| Feb 25, 2020 at 2:55 | comment | added | Wheat Wizard Mod | @JonathanFrech One example would be Haskell which has unbound integers. Many languages have these. But even C can output an arbitrary stream of digits as a string. | |
| Feb 25, 2020 at 2:49 | comment | added | Jonathan Frech | I see ... What language would be viable in the sense that it allows for an implementation of a program with codomain \$\mathbb{Z}\$? | |
| Feb 25, 2020 at 2:35 | comment | added | Wheat Wizard Mod | @JonathanFrech To answer your first question, you can get unbounded entropy by repeatedly sampling the time (variations in compute time can make successive samples effectively independent of the initial) but taking a single time prevents you from sampling more than once. To answer your second question since C has bounded integer size there are integers that that will not output making it not a valid answer. | |
| Feb 25, 2020 at 2:22 | comment | added | Jonathan Frech | What score would this C program get? | |
| Feb 25, 2020 at 2:20 | comment | added | Jonathan Frech | However, does this problem not already exist, just hidden? Most implementations use an RNG with finite, not provably controllable entropy assumed to be distributed uniformly. | |
| Feb 25, 2020 at 2:13 | comment | added | Wheat Wizard Mod | @JonathanFrech A time parameter has a few issues. It is basically just an integer so we need it to be unbounded (otherwise only a finite number of outputs can be attained) and on top of that we need to specify a distribution on the input for a distribution on the output to make sense. There is not really a clear best distribution for time, even harder is one that is not exploitable to the extent it trivializes the challenge. | |
| Feb 25, 2020 at 1:40 | comment | added | Jonathan Frech | @PostRockGarfHunter You could possibly allow a time parameter and require "reasonable" time independence of the solution. | |
| Feb 25, 2020 at 1:35 | comment | added | Wheat Wizard Mod | @JonathanFrech I would like to but it is a little difficult for entropy to make it work. If you have a concrete idea for how I might fairly introduce entropy, I am open to ideas. | |
| Feb 25, 2020 at 1:16 | comment | added | Jonathan Frech | I would allow for some entropy input to give pure languages a chance. | |
| Feb 14, 2020 at 16:07 | comment | added | xnor | Let us continue this discussion in chat. | |
| Feb 14, 2020 at 15:59 | comment | added | xnor | Maybe I'm missing something about loosening the distribution. I'm envisioning solutions in a golfing language to be about 7 bytes of "take a number N and make a flat random distribution based on it" and 93 bytes of "generate a huge N". For regular languages with a random() function, maybe split it 25 bytes / 75 bytes. Maybe I'm missing something, but I can't really see a viable strategy that isn't effectively the N-input challenge I suggested but with big number generation code copy-pasted in for N. Do you think there is? | |
| Feb 14, 2020 at 15:48 | comment | added | Wheat Wizard Mod | @xnor I think that I am pretty happy with the challenge as is. I prefer the current challenge over the suggested challenges since this one I feel has an incentive to loosen the distribution whereas I do not feel the others do. The big number generation is important, but I consider that as a more general technique. | |
| Feb 14, 2020 at 15:43 | comment | added | xnor | Alternatively, the "big number" N could be an input, and you stipulate that as N grows, the maximum chance of any integer to be chosen must approach zero. | |
| Feb 14, 2020 at 15:42 | comment | added | xnor | What do you think then about the challenge being just to generate a random integer so that each one has nonzero probability? I think that as is, the big number generation is so important to the score that it makes this a chameleon challenge. | |
| Feb 14, 2020 at 15:39 | history | edited | Wheat WizardMod | CC BY-SA 4.0 | Added a mathematical formulation |
| Feb 14, 2020 at 15:36 | comment | added | Wheat Wizard Mod | My thought is that there are a number of possible things that you would be able to use as the random part each with different drawbacks and strengths. In my mind the challenge is sort of coming up with the random part that best fits the language. There is definitely going to be a big number component to this that is important, but since that has mostly been explored in other challenges I would expect borrowing there. | |
| Feb 14, 2020 at 15:33 | comment | added | xnor | You're right, the byte limit stops you from actually getting arbitrarily close to 0. But, I expect there to be plenty of room in 100 bytes to stuff in some ginormous number bust-beaver style, at least in fairly compact languages. So, I think this challenge will mostly come down to "what's the biggest number you can express in 100-X bytes" where X is however many bytes the random part takes. | |
| Feb 14, 2020 at 15:28 | comment | added | Wheat Wizard Mod | @xnor There is a byte limit which stops "arbitrarily close" formulations. Did you miss this or am I missing what you are suggesting? | |
| Feb 14, 2020 at 14:45 | comment | added | xnor | This seems easy to score arbitrarily close to zero. For example, keep generating a random number from 1 to N until you get a 1, and count how many tries it takes before you succeed. To include negative outputs, pick a random sign. Make N large and the score tends to 0. | |
| Feb 14, 2020 at 14:07 | comment | added | Wheat Wizard Mod | @Lyxal It is mathematically impossible for all integers to have equal probability. | |
| Feb 14, 2020 at 12:08 | comment | added | simonalexander2005 | If all integers are equally probably, wouldn't that give a score of effectively 0? (probability of 1/Integer.Max for the "most frequently occuring", which could be any of the integers)? | |
| Feb 14, 2020 at 6:22 | comment | added | lyxal Mod | Of course, I may have missed the mark completely. Are we required to create our own scale of randomness (i.e. our own deviation from uniform randomness if that makes sense)? | |
| Feb 14, 2020 at 6:21 | comment | added | lyxal Mod | What happens if all integers are equally probable? Is that answer disqualified? Or does it score infinity? I also take it that there is a factor of randomness here, so perhaps you could mention that. | |
| Feb 14, 2020 at 5:01 | history | answered | Wheat WizardMod | CC BY-SA 4.0 |