jvecfor: Fast K-Nearest Neighbor Search for Single-Cell Analysis
Source:R/jvecfor-package.R
jvecfor-package.RdDrop-in replacement for BiocNeighbors::findKNN using the jvecfor
Java library (HNSW-DiskANN approximate and VP-tree exact methods). Achieves
approximately 2x speedup over Annoy-based search at n >= 50K cells.
Convenience wrappers delegate SNN/KNN graph construction to the bluster
package.
Main functions
fastFindKNNKNN search – returns index + distance matrices.
fastMakeSNNGraphKNN -> SNN graph via bluster.
fastMakeKNNGraphKNN -> KNN graph via bluster.
JvecforParamBiocNeighbors parameter class for drop-in integration with scran, scater, etc.
jvecfor_setupInstall a custom jvecfor JAR.
Options
jvecfor.verboseLogical. Enable Java/jvecfor progress logging globally. Default
FALSE.jvecfor.jarCharacter. Path to a custom jvecfor JAR file. Overrides the bundled JAR in
inst/java/.
Author
Maintainer: Anestis Gkanogiannis anestis@gkanogiannis.com (ORCID)