import nltk
nltk.download("words")| Corpus | Algorithm | Win rate |
|---|---|---|
| NLTK words | RandomVocabElimination | 0.72 |
| NLTK words | CharacterFrequencyVocabElimination | 0.14 |
| NLTK words | RandomUniqueCharVocabElimination | 0.83 |
| NLTK words | RandomCharFreqUniqueVocabElimination | 0.80 |
| NLTK words | RandomVocabElimination+ | 0.91 |
| NLTK words | RandomUniqueCharVocabElimination+ | 0.94 |
| NLTK words | InfoSeekingVocabElimination (v1) | 0.40 |
CharacterFrequencyVocabElimination picks "arara" first in NLTK words corpus which
New strategy: pick the word which you think will eliminate the most options
- This includes words that we know can't be the target but might give us lots of information
Consider this failure case:
Answer: bulky
a w f u l
c l o u d
s l u i t
g l u m p
l u r k y
h u l k y
Heuristics for what words are "valuable":
- Contains all unique letters
- Doesn't repeat any (char, position) that we're used before and we know are wrong
- Contains lots of letters we haven't tried yet
- Contains letters that we know are in the word but haven't positioned yet
Other heuristics based on this idea:
- If there's only one letter remaining to guess
- e.g. * u s h y, could be "bushy", "cushy" or "mushy"
- then it might be worth submitting "climb" as it will find out which of c, m or b is in the remaining slot
More failure cases:
Answer: douse
b i l g e
c o p s e
h o r s e
y o u s e
t o u s e
m o u s e
Answer: varve
m o i s e
a r u k e
c a r t e
p a r g e
l a r v e
w a r v e