There are a few types of indexes that are expensive to compute, inncluding tfidf for cards, and card_rank.
These need to be updated every time that a card is updated, but aren't THAT expensive. tfidf can change more slowly than card_rank may.
For both of these ideally every client doens't have to constantly recompute.
In addition, card_rank should be used for sorts, but sort functions only take in a card. Ideally there's some way to mix in data into a card like card_rank. Or I guess just make it so sorts also take card-rank.
Note that the pagerank library is 314kb at least to include. Ideally it wouldn't be included unless the user explicitly sorted by card-rank.
Another approach is to have an indexes collection, one for tfidf and one for card-rank that are fetched and kept updated. But that could be a lot of data to update, and it still would need to be merged in somewhere into the cards to be used for sorting.
jkomoros
commented
4 years ago
Originally captured in #151.
There are a few types of indexes that are expensive to compute, inncluding tfidf for cards, and card_rank.
These need to be updated every time that a card is updated, but aren't THAT expensive. tfidf can change more slowly than card_rank may.
For both of these ideally every client doens't have to constantly recompute.
In addition, card_rank should be used for sorts, but sort functions only take in a card. Ideally there's some way to mix in data into a card like card_rank. Or I guess just make it so sorts also take card-rank.
Note that the pagerank library is 314kb at least to include. Ideally it wouldn't be included unless the user explicitly sorted by card-rank.
Another approach is to have an indexes collection, one for tfidf and one for card-rank that are fetched and kept updated. But that could be a lot of data to update, and it still would need to be merged in somewhere into the cards to be used for sorting.