A package for precise approximative nearest neighbor search in more than just euclidean space.
Its only exported function find_knn computes the k nearest neighbors of the rows of the query matrix in the data matrix. If no query matrix is passed, the nearest neighbors for all rows in the data will be returned (i.e. data will be used as query).
find_knn(
data, k, ...,
query = NULL,
distance = c("euclidean", "cosine", "rankcor"),
sym = TRUE)
The result will be a list containing
index, a nrow(query) × k integer matrix containing the row indices into data that are the nearest neighbors.
dist, a nrow(query) × k double matrix containing the distances to those neighbors.
dist_mat, a nrow(query) × nrow(data) a Matrix::dSparseMatrix, generic if !sym or !is.null(query), and symmetric if sym and is.null(query). Zeros in this matrix mean “not a knn”, and if sym is set, the matrix will be post processed to be symmetric.
(Without post processing, the matrix will likely be asymmetric as r1∈kNN(r2) does not imply r2∈knn(r1))
This package was separated from destiny as it might prove helpful in other contexts. It provides more distance metrics than FNN and is more precise than RcppHNSW, but slower than both.
If anyone knows a faster and similarly precise kNN search in cosine (=rank correlation) space, please tell me!
RetroSearch is an open source project built by @garambo | Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4