Crate flow_pacmap 

flow-pacmap

First-party implementation of PaCMAP (Pairwise Controlled Manifold Approximation Projection) from Wang et al. 2021 (JMLR 22, Algorithm 1).

Designed for large-n flow cytometry data:

• No ndarray version conflicts — pure &[f32] / Vec<[f32;2]> API
• PCA via faer SVD on the d×d covariance matrix (O(n·d²), no large intermediates)
• HNSW KNN via usearch (C++ FFI, hardware SIMD, optional f16 quantization)
• All pair counts use checked_mul; no debug-mode overflow panics
• Progress reporting via mpsc::Sender<PaCMAPProgress> (per phase + every 10 iters)
• Cancellation via Arc<AtomicBool>

Re-exports

pub use config::DistanceMetric;

pub use config::HnswParams;

pub use config::Init;

pub use config::KnnMethod;

pub use config::PaCMAPConfig;

pub use config::Quantization;

pub use error::PaCMAPError;

Modules

adam

Adam optimizer step for the 2-D embedding.

config

Configuration types for a PaCMAP embedding run.

error

Error types for flow-pacmap.

gradient

PaCMAP loss gradient computation (Algorithm 1, Wang et al. 2021).

knn

K-nearest-neighbour search for PaCMAP graph construction.

pairs

Near, mid-near, and further pair construction from Algorithm 1.

pca

PCA initialisation via faer SVD on the d×d covariance matrix.

weights

Three-phase weight schedule from Algorithm 1 of Wang et al. 2021.

Structs

PaCMAPProgress

Progress event emitted during optimization.

Functions

fit_transform

Embed n × d row-major f32 data into 2 dimensions.