Skip to content

Top level

Module-level helpers, re-exported from the compiled core.

Vector search for single-cell and computational biology, in Rust.

Every index is a scikit-learn style estimator: parameters go in the constructor, data goes into fit, results come out of kneighbors.

>>> import numpy as np, ann_search as ann
>>> X = np.random.default_rng(0).standard_normal((5000, 50)).astype("float32")
>>> index = ann.HnswIndex(n_neighbors=15, metric="cosine").fit(X)
>>> distances, indices = index.kneighbors()        # self-kNN, fast path
>>> distances, indices = index.kneighbors(X[:100]) # cross-set

kneighbors returns distances first, matching scikit-learn and FAISS. A query that came back with fewer than k neighbours is padded with -1 indices and infinite distances, and kneighbors_graph drops those slots.

Indices are immutable. There's no incremental add, so rebuild instead.

AnnSearchError

Bases: Exception

Base class for every error raised by the Rust core.

IndexIoError

Bases: AnnSearchError

An index bundle is missing, truncated, or of the wrong kind or dtype.

gpu_available builtin 

gpu_available() -> bool

Whether the GPU indices can be used here.

Returns

True only when this build has the gpu feature and wgpu resolves an adapter. Safe to call on any machine.

Note

Acquiring a client panics rather than erroring when no adapter is found, so the probe catches it. That is sound here because the release profile is pinned to panic = "unwind", which pyo3 requires anyway. The panic hook is silenced for the duration, otherwise merely asking the question prints a backtrace to stderr.

num_threads builtin 

num_threads() -> int

Number of threads currently available for index builds and queries.

Returns

The configured pool's size, or rayon's global pool size when no override is in place.

set_num_threads builtin 

set_num_threads(n: int) -> None

Set the number of threads used for index builds and queries.

Rayon worker threads do not survive fork, so a multiprocessing child using the default fork start method on Linux will hang if it touches the pool. Use the spawn start method there.

Params
  • n - Thread count. 0 clears the override and returns to rayon's global pool, which honours RAYON_NUM_THREADS.
Returns

Nothing, or a RuntimeError if the pool could not be built.