Crate flow_knn 

flow-knn

Algorithm-agnostic k-nearest-neighbour graphs for large-n scientific data.

Build once with compute_knn, reuse the KnnGraph across embedders (PaCMAP, UMAP, …) without recomputing neighbours.

Structs

HnswParams

Quality / memory trade-off for HNSW indices (usearch or ann-search-rs).

KnnGraph

Portable k-nearest-neighbour graph: per-point indices and distances only.

NeighborList

Neighbours of a single query point.

PerfRecord

One timed cell in the (n, d, method) performance matrix.

RecommendOpts

Preferences that bias automatic method choice.

Enums

DistanceMetric

Distance metric for the KNN graph.

KnnError

KnnMethod

Approximate / exact nearest-neighbour method.

Quantization

Vector quantization for usearch HNSW storage.

Functions

builtin_matrix

Built-in snapshot used when JSONL parse fails.

compute_knn

Compute k nearest neighbours for all n points in row-major data.

exact_knn

Exact k-NN for all n points in row-major data.

load_matrix

Load matrix: prefer path if readable, else the shipped JSONL / builtin_matrix.

parse_matrix_jsonl

Parse JSONL text into records (skips blank / comment lines).

read_knn_graph

Read a KnnGraph previously written by write_knn_graph.

recommend_method

Recommend a KnnMethod for n points of dimension d.

recommend_method_with_matrix

Same as recommend_method but with an explicit matrix (tests / custom datasets).

write_knn_graph

Write a KnnGraph to path (creates/overwrites the file).