ayyy PR
Semantle #17 in 80 guesses. semantle.novalis.org
Conversation
I solved Semantle #17 in 65 guesses. semantle.novalis.org
2
1
nice! i think sub-50 is possible here if you're more conservative with your guesses than i was, hoping to reach that milestone in the next few days and maybe try to maintain a long-run sub-100 average in the meantime
1
1
I still don't yet have a feel for how closeness is calculated. I'm wondering if it's worth entering generic words like "thing", "action", etc.
1
1
the training set here was newspaper articles so you want words that appear in a lot of those. word2vec is trained on a prediction task involving predicting a word from the surrounding words in the corpus
1
1
so you want to think about something like, given a word, the following 2-step inference:
1) which sentences in which kinds of newspaper articles is this word likely to appear in
2) which other words are likely to appear in similar sentences in similar articles
sbe gbqnl'f jbeq v tbg obql cnegf naq gura vawhel, fb v jnf guvaxvat nobhg fragraprf yvxr "gur sbbgonyy cynlre jnf vawherq va uvf ____" naq gelvat gb svyy va gur oynax. gura v tbg obar jbeqf, thrffrq zneebj, gura thrffrq genafcynag ba n juvz juvpu jnf 898/1000
1
1
Lrnu V qvq obql cnegf, gura "nangbzl" juvpu gbbx zr va n zrqvpny qverpgvba, gura "qbpgbe" juvpu jnf 997/1000.
1