#hashing

Question 1: Single choice

English has about $3 \times 10^8$ possible character 6-grams, and fastText represents every word by the sum of the vectors of its 3- to 6-grams. How does fastText keep its table of n-gram vectors bounded?

It hashes every n-gram into one of $2 \times 10^6$ buckets, so colliding n-grams share a vector1
It keeps only the 3-grams, since there are few enough of those to store2
It stores vectors only for the n-grams seen at least 100 times in the corpus and drops all others3
It stores no n-gram vectors and computes them from character embeddings at run time4

Timed test