Closed ctb closed 3 years ago
yes, I think we should. :)
given the stuff going on with greyhound, we are going to ignore performance in this paper (beyond implying that it's acceptable, 'cause here are the results).
(and in fact we are removing that entire section as part of shift to #10)
In Results section Scaled MinHash sketches support efficient indexing for large-scale containment queries, tbl:search-runtime shows runtime for similarity search. Two thoughts --
first, these are surprisingly slow :(. second, these are for similarity, not containment.
My experience with containment and gather (which uses containment) is that these are pretty fast operations; I rather rarely use similarity. Moreover, the whole paper is more focused on containment than similarity anyway.
Should we refocus this benchmark on containment?