"Guess My Word" Gurus
June 1, 2013 11:22 AM Subscribe
not_on_display introduced us to this game back in December : FPP. I do okay, but usually need 11-18 tries to solve it. What strategies do those of you with the lower numbers use?
My general pattern for words before Ma is - ma, gab, lab, habit, ice, etc. and then narrow it down from there. I use the same strategy for words after Ma as well. It works, eventually, but I'd love to improve.
My general pattern for words before Ma is - ma, gab, lab, habit, ice, etc. and then narrow it down from there. I use the same strategy for words after Ma as well. It works, eventually, but I'd love to improve.
I think the correct strategy is to split the difference. Pick the middle letter (M?) and then branch off from there. If it says before, you go to "F", if it says after, you go to "S". If you keep splitting the target group of words in two, you narrow down to the correct letter in no more than 5 steps, I think, and then you narrow down the words the same way. It probably helps if you were a spelling bee person.
Probably best to choose the first dictionary word with that letter, so you don't get any false positives. If you choose "mad" and the word is "mabe", you can get fooled that the word doesn't start with M when it does.
posted by gjc at 12:12 PM on June 1, 2013
Probably best to choose the first dictionary word with that letter, so you don't get any false positives. If you choose "mad" and the word is "mabe", you can get fooled that the word doesn't start with M when it does.
posted by gjc at 12:12 PM on June 1, 2013
How much do you want to cheat? You could probably improve the straight binary-ish search by picking wisely. For instance in the simple /usr/share/dict/words on my system the greatest number of words start with:
h 3804
e 3851
f 4127
t 5085
r 5356
d 5823
a 5895
m 5922
b 6068
p 7659
c 9521
s 11327
and second letters go:
u 7734
r 8324
i 9960
o 14001
e 14399
a 15180
I'm sure you could narrow the search down faster by picking words more like you were searching through a physical dictionary of words vs searching some imaginary (a-z)+ space of random strings.
posted by zengargoyle at 12:22 PM on June 1, 2013
h 3804
e 3851
f 4127
t 5085
r 5356
d 5823
a 5895
m 5922
b 6068
p 7659
c 9521
s 11327
and second letters go:
u 7734
r 8324
i 9960
o 14001
e 14399
a 15180
I'm sure you could narrow the search down faster by picking words more like you were searching through a physical dictionary of words vs searching some imaginary (a-z)+ space of random strings.
posted by zengargoyle at 12:22 PM on June 1, 2013
The best non-cheating strategy is a binary search, which gjc is almost doing. The difference between gjc's strategy and the optimal strategy is to pick the middle word in the dictionary, not the middle letter. They say in the FAQ which dictionary they use, so you can use it to determine that (I'm making this example up) the middle word is "metal", rather than the middle letter being "m". In fact, the middle word might not even begin with m or n, due to the unequal distribution of words.
Then after it tells you "metal" is low, you can use the dictionary to determine that the middle word between "metal" and the end of the words is, say, "restful", and choose that. And so on.
No other strategy will do better, on average, except for cheating strategies (which some people clearly use).
posted by Flunkie at 12:25 PM on June 1, 2013 [1 favorite]
Then after it tells you "metal" is low, you can use the dictionary to determine that the middle word between "metal" and the end of the words is, say, "restful", and choose that. And so on.
No other strategy will do better, on average, except for cheating strategies (which some people clearly use).
posted by Flunkie at 12:25 PM on June 1, 2013 [1 favorite]
And I guess I should say that I'm aware that some people would consider that cheating. But it clearly would not be cheating if you had the dictionary memorized, and I think it's thus at least a different type of beast than cheating in the sense of looking up the word for the day and then playing again (perhaps on a different computer) and selecting that word as your first guess. Which some people clearly do.
posted by Flunkie at 12:27 PM on June 1, 2013
posted by Flunkie at 12:27 PM on June 1, 2013
Since I've been called out in the first comment, I will try to give this some more thought, but since my average is 15.75, I'd first have to point out that 11-18 is a pretty good range.
Also if you enjoy seeing your name near the top of the leaderboard, play early.
posted by MCMikeNamara at 2:20 PM on June 1, 2013
Also if you enjoy seeing your name near the top of the leaderboard, play early.
posted by MCMikeNamara at 2:20 PM on June 1, 2013
You can see from the overall leaderboards (100+ games) that the overall guess leader averages 12.66 guesses, and the overall time leader does it in 1:05. My average is 17.09/2:50, so I won't offer my strategy! :)
posted by not_on_display at 2:37 PM on June 1, 2013
posted by not_on_display at 2:37 PM on June 1, 2013
Best answer: The mathematically best strategy is to do a binary search, as others have said (and you're not terribly far from that). One improvement is to guess likely words; while anything in the scrabble dictionary is fair game, they are trying to have fun and aren't going to choose something short and simple like CAT or obscure like XU or UXORICIDE as the target word. Of their last 50 words, 88% were between 5-9 letters, with only one 4 letter word, two 11 letter words and one 14 letter word. So guessing ma, gab, lab and ice mean you are making guesses that are very unlikely to actually be the answer.
I took a word list (2of12, which is a smaller list of 40K words but still had 48 of the last 50 words in the game) and filtered out the 5 to 9 letter words. The optimal opening strategy based on that set of words is:
1st pick: literary
2nd pick: diurnally or sands
3rd pick: calamine, godmother, patterned, tease
4th pick: bailable, copious, farrier, important, muskrat, queerness, solicitor, unseal
5th pick: ancestral, bloop, chromatic, dealt, emotion, forsworn, hectogram, jumper, megabucks, oneself, pommel, repent, shill, strange, treetop, warpath
posted by Homeboy Trouble at 3:23 PM on June 1, 2013 [2 favorites]
I took a word list (2of12, which is a smaller list of 40K words but still had 48 of the last 50 words in the game) and filtered out the 5 to 9 letter words. The optimal opening strategy based on that set of words is:
1st pick: literary
2nd pick: diurnally or sands
3rd pick: calamine, godmother, patterned, tease
4th pick: bailable, copious, farrier, important, muskrat, queerness, solicitor, unseal
5th pick: ancestral, bloop, chromatic, dealt, emotion, forsworn, hectogram, jumper, megabucks, oneself, pommel, repent, shill, strange, treetop, warpath
posted by Homeboy Trouble at 3:23 PM on June 1, 2013 [2 favorites]
Response by poster: Thanks all. I'm going to try all of these ideas. Happy playing!
posted by 1066 at 11:33 PM on June 1, 2013
posted by 1066 at 11:33 PM on June 1, 2013
This thread is closed to new comments.
You can study the guess histories of people who do well to see what their strategies are. I use pretty much the same short words every time in the beginning of the game to maximize my speed, then once I get it narrowed down to the correct letter, switch to harder words to maximize my chances of hitting on a guess while I'm still narrowing it down further. But I notice that the really good players tend to use hard words all the way through. Probably because they're just plain old smarter. And faster typists.
I'm not that great at it though. The person whose advice we want is MCMikeNamara. Even though he cheated on fescue. :)
posted by HotToddy at 12:06 PM on June 1, 2013 [1 favorite]