Crate ferrolearn_decomp 

ferrolearn-decomp

Dimensionality reduction and matrix decomposition for the ferrolearn machine learning framework.

This crate provides PCA, TruncatedSVD, NMF, Kernel PCA, and manifold learning methods that follow the ferrolearn Fit/Transform trait pattern.

Algorithms

• PCA — Principal Component Analysis. Centres data and projects onto the directions of maximum variance.
• TruncatedSVD — Truncated Singular Value Decomposition using the randomized algorithm. Does not centre data, making it suitable for sparse inputs.
• NMF — Non-negative Matrix Factorization. Decomposes a non-negative matrix X into W * H where both factors are non-negative.
• KernelPCA — Kernel PCA. Non-linear dimensionality reduction via a kernel-induced feature space.
• MDS — Classical Multidimensional Scaling. Embeds data preserving pairwise distances.
• Isomap — Isometric Mapping. Non-linear dimensionality reduction via geodesic distances on a kNN graph.
• SpectralEmbedding — Laplacian Eigenmaps. Non-linear dimensionality reduction via the normalised graph Laplacian.
• LLE — Locally Linear Embedding. Non-linear dimensionality reduction preserving local reconstruction weights.
• Tsne — t-distributed Stochastic Neighbor Embedding. Non-linear dimensionality reduction using Barnes-Hut approximation.
• Umap — Uniform Manifold Approximation and Projection. Fast non-linear dimensionality reduction based on topological data analysis.
• LatentDirichletAllocation — Latent Dirichlet Allocation topic model. Discovers latent topics in document-term matrices.
• DictionaryLearning — Sparse coding with a learned dictionary.

Pipeline Integration

PCA<f64>, TruncatedSVD<f64>, NMF<f64>, and KernelPCA<f64> all implement PipelineTransformer so they can be used as transformer steps in a Pipeline.

Examples

use ferrolearn_decomp::PCA;
use ferrolearn_core::traits::{Fit, Transform};
use ndarray::array;

let pca = PCA::<f64>::new(1);
let x = array![[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]];
let fitted = pca.fit(&x, &()).unwrap();
let projected = fitted.transform(&x).unwrap();
assert_eq!(projected.ncols(), 1);

Re-exports

pub use covariance::EllipticEnvelope;

pub use covariance::EmpiricalCovariance;

pub use covariance::FittedCovariance;

pub use covariance::FittedEllipticEnvelope;

pub use covariance::FittedLedoitWolf;

pub use covariance::FittedMinCovDet;

pub use covariance::FittedOAS;

pub use covariance::LedoitWolf;

pub use covariance::MinCovDet;

pub use covariance::ShrunkCovariance;

pub use covariance::OAS;

pub use cross_decomposition::CCA;

pub use cross_decomposition::FittedCCA;

pub use cross_decomposition::FittedPLSCanonical;

pub use cross_decomposition::FittedPLSRegression;

pub use cross_decomposition::FittedPLSSVD;

pub use cross_decomposition::PLSCanonical;

pub use cross_decomposition::PLSRegression;

pub use cross_decomposition::PLSSVD;

pub use dictionary_learning::DictFitAlgorithm;

pub use dictionary_learning::DictTransformAlgorithm;

pub use dictionary_learning::DictionaryLearning;

pub use dictionary_learning::FittedDictionaryLearning;

pub use factor_analysis::FactorAnalysis;

pub use factor_analysis::FittedFactorAnalysis;

pub use fast_ica::Algorithm;

pub use fast_ica::FastICA;

pub use fast_ica::FittedFastICA;

pub use fast_ica::NonLinearity;

pub use incremental_pca::FittedIncrementalPCA;

pub use incremental_pca::IncrementalPCA;

pub use isomap::FittedIsomap;

pub use isomap::Isomap;

pub use kernel_pca::FittedKernelPCA;

pub use kernel_pca::Kernel;

pub use kernel_pca::KernelPCA;

pub use lda_topic::FittedLatentDirichletAllocation;

pub use lda_topic::LatentDirichletAllocation;

pub use lda_topic::LdaLearningMethod;

pub use lle::FittedLLE;

pub use lle::LLE;

pub use mds::Dissimilarity;

pub use mds::FittedMDS;

pub use mds::MDS;

pub use nmf::FittedNMF;

pub use nmf::NMF;

pub use nmf::NMFInit;

pub use nmf::NMFSolver;

pub use pca::FittedPCA;

pub use pca::PCA;

pub use spectral_embedding::Affinity;

pub use spectral_embedding::FittedSpectralEmbedding;

pub use spectral_embedding::SpectralEmbedding;

pub use truncated_svd::FittedTruncatedSVD;

pub use truncated_svd::TruncatedSVD;

pub use tsne::FittedTsne;

pub use tsne::Tsne;

pub use umap::FittedUmap;

pub use umap::Umap;

pub use umap::UmapMetric;

Modules

covariance

Covariance estimation.

cross_decomposition

Cross-decomposition methods: PLS, CCA, and PLSSVD.

dictionary_learning

Dictionary Learning.

factor_analysis

Factor Analysis (FA) via the EM algorithm.

fast_ica

Fast Independent Component Analysis (FastICA).

incremental_pca

Incremental Principal Component Analysis (IncrementalPCA).

isomap

Isomap (Isometric Mapping).

kernel_pca

Kernel Principal Component Analysis (Kernel PCA).

lda_topic

Latent Dirichlet Allocation (LDA) topic model.

lle

Locally Linear Embedding (LLE).

mds

Multidimensional Scaling (MDS).

nmf

Non-negative Matrix Factorization (NMF).

pca

Principal Component Analysis (PCA).

spectral_embedding

Spectral Embedding (Laplacian Eigenmaps).

truncated_svd

Truncated Singular Value Decomposition.

tsne

t-distributed Stochastic Neighbor Embedding (t-SNE).

umap

Uniform Manifold Approximation and Projection (UMAP).