Build an Annoy index from a bigmemory::big.matrix — annoy_build

Stream the rows of a reference bigmemory::big.matrix into an on-disk Annoy index and write a small sidecar metadata file next to it. The returned bigannoy_index can be reopened later with annoy_open_index().

Usage

annoy_build_bigmatrix(
  x,
  path,
  n_trees = 50L,
  metric = "euclidean",
  seed = NULL,
  build_threads = -1L,
  block_size = annoy_default_block_size(),
  metadata_path = NULL,
  load_mode = "lazy"
)

Arguments

x: A bigmemory::big.matrix or an external pointer referencing the reference matrix.
path: File path where the Annoy index should be written.
n_trees: Number of Annoy trees to build.
metric: Distance metric. bigANNOY v2 supports "euclidean", "angular", "manhattan", and "dot".
seed: Optional positive integer seed used to initialize Annoy's build RNG.
build_threads: Build-thread setting passed to Annoy's native backend. Use -1L for Annoy's default.
block_size: Number of rows processed per streamed block while building the index.
metadata_path: Optional path for the sidecar metadata file. Defaults to paste0(path, ".meta").
load_mode: Whether to keep the returned index metadata-only until first search ("lazy") or eagerly load a live index handle immediately ("eager").

Value

A bigannoy_index object describing the persisted Annoy index.