lshSizing
lshSizing(
threshold,k,weights?):LshSizing
Defined in: packages/distillate/src/minhash/sizing.ts:91
Bands and rows for an LSH index over k-slot signatures, chosen as
datasketch does: a pair at similarity s becomes a candidate with
probability 1 - (1 - s^rows)^bands, and the pair minimising
weights[0] * falsePositive + weights[1] * falseNegative around
threshold wins. Balanced weights put the curve’s midpoint near the
threshold, so pairs just above it are found most of the time, not always.
The search tries every split of k, so its cost grows with k: about
10 ms at 128 and a few hundred ms at 4096.
Parameters
Section titled “Parameters”threshold
Section titled “threshold”number
Jaccard similarity that should make a pair a candidate.
number
Slots in each signature.
weights?
Section titled “weights?”readonly [number, number] = ...
Cost of a false positive and of a false negative; sum to 1.