Crate manifolds_rs 

Dimensionality reduction algorithms including UMAP, t-SNE, PHATE, Diffusion Maps and PacMAP.

Provides both standard and approximate nearest-neighbour-based graph construction, multiple optimisers, and (optionally) parametric UMAP via a neural network encoder.

Additionally, optional GPU-accelerated versions (in terms of kNN search) can be used when the right feature flags are active.

Modules

data

Module containing the initial data to generate the embedding. This includes generation of the kNN graphs, modifications of these prior to usage in the different methods, structures to store the data (+ helpers) and also synthetic data.

errors

Errors that can occur in manifolds-rs

prelude

Re-exports of commonly used types and traits for convenient glob importing.

training

Contains all of the different optimiser to fit PHATE, tSNE and UMAP.

utils

Contains key utility functions, like BH trees, FFT for tSNE, diffusion and potentials for PHATE, sparse operations across different functions and also macros

Macros

assert_same_len

Assertion that all vectors have the same length.

Structs

DiffusionMapsParams

Parameters for classical diffusion maps (Coifman & Lafon, 2006).

PacmapParams

Parameters for PaCMAP dimensionality reduction.

PhateParams

PHATE parameters

TsneParams

Main configuration for t-SNE dimensionality reduction

UmapParams

Main Config structure with all of the possible sub configurations

Enums

DiffusionMapsOperator

Full or landmark diffusion maps operator.

Functions

construct_diffusion_maps_operator

Build the diffusion maps operator from raw data.

construct_phate_diffusion

Build the PHATE diffusion operator from high-dimensional data

construct_tsne_graph

Construct affinity graph for t-SNE from high-dimensional data

construct_umap_graph

Helper function to generate the UMAP graph

diffusion_maps

Run diffusion maps end-to-end.

pacmap

Run PaCMAP dimensionality reduction.

phate

Run PHATE dimensionality reduction

tsne

Run t-SNE dimensionality reduction

umap

Run UMAP dimensionality reduction

Type Aliases

PreComputedKnn

Type for the pre-computed kNN

TsneGraph

Result of the tSNE graph