kmizu / sketches Goto Github PK

import atrox.sketch._

val sets: Seq[Set[Int]] = loadMyData()

val (bands, hashes) = LSH.pickHashesAndBands(threshold = 0.5, maxHashes = 64)
val minhash = MinHash(sets, hashes)
val lsh     = LSH(minhash, bands)

val cfg = LSHCfg(maxResults = 50)

for (Sim(idx1, idx2, estimate, similarity) <- lsh.withConfig(cfg).allSimilarItems(minEst = 0.5)) {
  println(s"similarity between item $idx1 and $idx2 is estimated to $estimate")
}

lsh.withConfig(LSHCfg(
  // Return only the 100 most relevant results.
  // It's strongly recommended to use this option.
  maxResults = 100,

  // Perform similarity search in parallel.
  parallel = true,

  // Use as much memory as needed. This leads to faster bulk queries but
  // might need to store the complete result set in memory.
  compact = false,

  // Skip anomalously large buckets. This speeds things up quite a bit.
  maxBucketSize = sets.size / 10
))

lsh.similarItems(idx)
lsh.similarIndexes(idx)
lsh.allSimilarItems()
lsh.allSimilarIndexes()