Post by luvrhino in Working out process

Viewing post in Working out process

It's all pretty much a binary search of trying to divide the candidates in half with half lower and half higher. I occasionally tweak that when I think certain candidates are more or less likely or that if I use a certain guess I will prompt a (yyy - X) is a square number or (yyy - X) is prime clue. That's the same logic of why we start around 500.

Again, it's important that it's dividing the candidates in two and not the remaining numbers. I describe that above when I discuss what I do with "x > 499; (x-499) is a perfect square," Guessing 755 (i.e., 499 + 16²) would be the closest to the midpoint of 499 and 999. The remaining candidates are 499 + 1² through 499 + 22². That suggests using 11². However, because I think 4², 8², 9², 16², and possibly 1² would all have generated different clues (e.g., "(x - 499) is a perfect 6th power" for 563 = 499 + 8²), I think 13² is probably best as it divides the remaining choices in evenly...8 above and 8 lower.

13² has a secret bonus of possibly triggering "x < 668; (668 - x) is a perfect square" if the solution were 524 or 643, thanks to the Pythagorean theorem and the 5-12-13 triangle. Now that I think of it, 12² is even better for this since you'd have a 5-12-13, 9-12-15, and 12-16-20 triangles all as options. "x > 643; (x -643) is a perfect square" for 668 or 724 and "x > 643; (x -643) is a perfect 8th power" for 899.

While I (obviously) tend to overthink things, this was the first time I noticed those 12² options. I rarely anticipate the "(x + yyy) is a perfect square" type clues, though I love it when I get them.

Similarly, I don't necessarily go through listing out all the candidates after the first clue, especially if I start with a mediocre-to-bad clue like "x > 499; (x-499)'s largest prime divisor is 5." In that case, I'll just guess where the midpoint among candidates will be, use something like 679 (= 499 + 2² * 3² * 5) and hope to get a different clue next. I wouldn't be surprised if 659 or even 619 were better, here, but I can't be bothered.

I hope that makes sense as to what my thought process is on subsequent guesses and what kinds of things one could consider if you wanted to play optimally. I do try to strike a balance in how much time to spend to try playing optimally vs. getting the thing done more quickly.

luvrhino1 year ago (1 edit) (+1)

Today (10/7) is one where using 499 is much better than your originally proposed method. I managed to get it in two. Starting with (512, 729) leaves 14 candidates, which isn't great. Depending on the clue, it is possibly better than if it didn't have a single 3 factor, only a single 2.

I think at that point using the 8th biggest would make sense, which is 870 assuming I'm counting correctly. That helps because if it returns with the largest divisor is 30, you'd know it's 750 or 930. Otherwise, since it's "x < 870; largest divisor is 6" you know it's not 750, leaving 6 candidates.

Using the 4th biggest, it's 822, which would leave 3 candidates. That is fine with two guesses remaining as you guess the middle one and go above or below if that's not correct. Instead, I got the "x<822; (822-x) is a square." Clearly (822 - x) = 6² -> x = 786...which was the middle of the 3 remaining, anyway.

With the benefit of hindsight, after the 3rd guess, I should have eliminated 834 because would have given the "is a square" clue instead of "largest divisor is 6." That would have left 5 candidates: 762, 786, 798, 822, 858. Choosing the middle one, means you'd either get it correct, get the "is a square clue" if it's 762, or get the "largest divisor is 6," if the solution is 786, 822, or 858. If it were 822 or 858 you're left with a coinflip to see if you get it in 5 or 6 guesses.

By contrast, with 499, my clue was "(x-499)'s largest prime divisor is 41." That leaves 11 or 12 candidates, since 540 might have generated a less useful (x-499) is prime clue. The 6th of the 11 or 7th of the 12 candidates is 499 + 7 * 41 = 499 + 287 = 786, which is the solution. That's luck, but it narrowed it down further in one guess than what 512/729 did in 2.

Anyway, I enjoy this kind of post-mortem analysis. It's fun to figure out what the decision tree should have been.

Kinwoods1 year ago

That's fair enough. Thanks for the tips. I didn't have time for yesterday's xdle, but yeah, 14 candidates doesn't sound great. 786 is divisible by 2 and 3 only once, so that's a worst case scenario there using my usual strat.

Today's (10/8), however, is really funny to me, because 512 and 499 give the same clue, given that they are 13 apart from each other. The 19 or 20 candidates are both whittled down to the same 3.

However, if I'm understanding you correctly, it seems you're saying is that xdle has a priority list for the hints it gives. If so, then that's a big tell, because if I don't get "x and 512 largest divisor is 2", then I know that the number is odd. 512 would've thus been more useful here; 499 gives 19 candidates, and 512 gives 20, but the even numbers can be discarded without thought, and thus we only have 10 to choose from.

That doesn't mean that 512 is always better, and frankly, I wouldn't know how to most efficiently proceed (for today's xdle, my opening was (499, 243)) after getting that list. 243 seems good enough to split the list in half, although I'd be interested in knowing what you did.

luvrhino1 year ago (1 edit)

Having no 2 or 3 factors should be worse than a single 2 or 3. Having no 2 factors leaves roughly half the numbers as candidates. Being divisible by 2, but not 4, limits it to about a quarter.

I've never seen the code for how xdle works and how it chooses its clues. A strict priority list of what clues it prefers follows what I've seen. I do know it's deterministic and not random as the same puzzle returns the same clues if the input is the same. I am not aware of it not giving a "perfect square/cube/4th power...etc." clue when that's an option.

As for the 10/8 puzzle, I decided 20 * 13 was good enough, and got it in two clues. That was pure luck. I would have done the same if I started at 512. In that case, it switches to "x<252; (252-x) is prime," narrowing it down to just 239.

Getting the "(252-x) is prime" clue is what I'd expect, based on what I think the clue priority list is. Similarly, if I guess 303 today, it gives "(303-x) is a 6th power."

Understanding that there is this hierarchy of clues can help eliminate some clues. When I bother to look deeper as I have above, projecting what sequence of clues it would give one or two steps down the line can reduce the number of guesses.

Finally, I'll note that, unlike Dordle, for xdle, I only guess numbers that are possible solutions. I wouldn't do your (512, 243) scheme unless it fit. I don't make impossible guesses frequently in Dordle, but sometimes they're necessary to eliminate several letters at once.

Kinwoods1 year ago (1 edit)

Having no 2 or 3 factors should be worse than a single 2 or 3. Having no 2 factors leaves roughly half the numbers as candidates. Being divisible by 2, but not 4, limits it to about a quarter.

Sure, but in the context of 512, not getting the largest divisor hint means that we can rule out half of the candidates of whatever other clue it gives us. But that's not something I ought to count on, obviously (today (10/9) 512 is unhelpful).

I've never seen the code for how xdle works and how it chooses its clues. A strict priority list of what clues it prefers follows what I've seen. I do know it's deterministic and not random as the same puzzle returns the same clues if the input is the same.

Not that it's likely to frequently be useful, but what do you think the priority list is? To me, it's probably "largest divisor" > "is prime/nth power" > "largest prime divisor" [ETA: Guessing 0 or 1 gives the amount of prime factors x has, but I'm pretty sure that that's only for 0 and 1]

for xdle, I only guess numbers that are possible solutions.

Seems reasonable enough. I'm probably more used to having a multitude of options (before I switched to 499 as my start). Speaking of, if you guessed 701, you'd get "largest prime divisor is 19". Would you say it's better to guess 834 or 853? My intuition is 853 ('cause it's odd), but I'd like to know your thoughts.

luvrhino1 year ago

As I mentioned, I don't have it fully mapped out.

I think it puts (x - 499) is a perfect nth power as the highest priority. The other possibility is (x+499) is a perfect nth power. I haven't been able to test which one is higher since those would only both be available for the same number for specific values of x.

Using some prime numbers away from 910, I did get these clues just now:

(x+927)'s largest prime divisor is 167
(x+893)'s largest prime divisor is 601
(x+933)'s largest prime divisor is 97
(x+953)'s largest prime divisor is 23

It preferred those to (963-x) is prime. 910 + 963 = 1873 is a prime number. Same if I tried 957 as 1867 is prime.

I don't recall it not giving "x is a multiple of (499 - x)" when that is an option and it can't do the nth-power clue. It does that for 912, 913, 920, 923, 975, and other number today. It does prefer that to (499 - x) is prime clue.

My only other discovery just now is that if I am 1 away from the solution, the game breaks. Hitting <enter> will cause it to become unresponsive. I can reload the game and make a new guess, however. I will report that bug in a different thread. I don't think I had ever been one away before, so I don't know if this bug is new or not.

Kinwoods1 year ago (3 edits)

Interesting. Today's (spoilers ahead) was a number, and something I found is that, while it doesn't give "is prime" for direct prime numbers away, it does when it's more than 2 multiples away:

395 (376 + 19) gives "x+395",
414 (376 + 19*2) gives "largest divisor is 2"
433 (376 + 19*3), however, gives "433 - x is prime". As does 319 (376 - 19*3).

Tangentially, I'm pretty confident in saying that the "largest divisor" clue is the 2nd highest priority (checking any, bar the nth powers, even number today (Oct 10) gives you that clue).

Are you sure that guessing only possible answers is a good idea? I guessed (499, 253) and got the same hint, which was amusing, but also quite unhelpful. Perhaps I should've tried an even number?

My only other discovery just now is that if I am 1 away from the solution, the game breaks.

That certainly is an... interesting bug.

luvrhino1 year ago (1 edit) (+1)

For 470 (376 + 94), I get "x is a multiple of (470-x)." It says that rather than "largest divisor is 94." Same is true for 423 and 564.

752 yields this: "x<752; (752-x) is a power of x,"

That is a new one for me. I had never tested double the solution before. Otherwise, that clue would be difficult to encounter, though you could have eventually starting with 512 & 243. Going the other direction, for 188, I get "x is a multiple of 188." Same for 94 and 47. I hadn't seen those either.

As for when I played for real, I did use (499, 212) because of a combination of eliminating 458 (as I'd get a "(499-x) is prime" clue) which made 212 the midpoint and because getting a "largest divisor clue" wouldn't be that bad at that point. That worked perfectly as it returned "x>212; x and 212's largest divisor is 4," which left 376 as the only possible solution.

If it were "largest divisor is 2," then I would have had 294 as the only solution more than 212 since 458 was already eliminated.

Going the other direction with x<212, the "largest divisor is 4" -> 048. "Largest divisor 2" -> 130. "(212-x) is prime" -> 171. "(212-x)'s largest prime divisor is 41" -> 007 or 089, though 089 might give "(x + 212) is prime."

I did not figure all that out live before I guessed. I just knew that getting a largest divisor clue would have been quite useful, which made 212 appealing beyond it being the midpoint among remaining candidates.

Kinwoods1 year ago (2 edits)

For 470 (376 + 94), I get "x is a multiple of (470-x)." It says that rather than "largest divisor is 94." Same is true for 423 and 564.

Ah, interesting. Guess I didn't check enough numbers, then.

752 yields this: "x<752; (752-x) is a power of x,"

I can see that hint being really useful, because that's basically an insta-win for anything above 249, which is 3/4 of the choices. But the likelihood of even getting that hint is quite rare.

Otherwise, that clue would be difficult to encounter, though you could have eventually starting with 512 & 243. Going the other direction, for 188, I get "x is a multiple of 188." Same for 94 and 47. I hadn't seen those either.

Hmm, after a bit of testing, I can't find any other hints. Think that's the lot of solutions then.

As for when I played for real, I did use (499, 212) because of a combination of eliminating 458 (as I'd get a "(499-x) is prime" clue) which made 212 the midpoint and because getting a "largest divisor clue" wouldn't be that bad at that point. That worked perfectly as it returned "x>212; x and 212's largest divisor is 4," which left 376 as the only possible solution.

Wish I were that lucky, cause I got the same clue for today (499, 233) and likewise for when I retried with (499, 200). Edit: I think I've been getting unlucky with your strat (I don't want to seem like I'm doubting your strategy, but genuinely (512, 243) would've given 3s for yesterday and today)

luvrhino1 year ago(+1)

Wish I were that lucky, cause I got the same clue for today (499, 233) and likewise for when I retried with (499, 200). Edit: I think I've been getting unlucky with your strat (I don't want to seem like I'm doubting your strategy, but genuinely (512, 243) would've given 3s for yesterday and today)

Today, I went (499, 246, 407). With 499, I had (499-x)'s largest prime divisor is 23, so I subtracted 11 * 23 = 253 to get 246, which is divisible by 2 and 3. Since I it returned the largest prime divisor is 23 clue again, I thought it was most likely to be 5 * 23 or 7 * 23 away. Otherwise, I'd get the "x and 246's largest divisor was 2" type clue. I was lucky enough to chose the latter.

However, yes, this is something that requires experimenting over time. You're correct that (512, 243) narrows today's down to a single solution, which makes it at least 0.5 guesses better than (499, 246) for today's number.

Except by accident, I have *not* played with having second guesses that aren't possible solutions. There may be certain clues where that's the optimal solution, but instinctively that feels wrong. However, that may also depend on what one's goal is...to minimize the average number of guesses or to minimize the number of losses. Those could yield slightly different strategies.

That does affect how I play Wordle, Dordle, and Quordle. Because I play against others, I used strategies that are optimized to beat my two friends at the slight increased risk of losing the game. Most notably, I stopped using ADIEU as a starter, because that made getting it in two extremely difficult and three less frequent than four.

For Dordle and Quordle, I do sometimes use words that aren't possible solutions to more efficiently eliminate letters. For example, I'd guess if the game returns -OUND, I might guess BUMPH to test four candidates for the first letter...even though that's British slang and would never be the solution. So, I'm open to something like your 243/729 second guess being best. I just feel it isn't for the reasons described elsewhere in the thread.

I will continue playing it both ways to compare.

View more in thread

itch.io

Viewing post in Working out process