Module: aprender::decomposition

Public module of the aprender-core crate.

Source

crates/aprender-core/src/decomposition.rs or directory.

Example

use aprender::decomposition::ICA;
// See `cargo doc -p aprender-core --open` for full API reference.

Module summary

aprender::decomposition houses matrix-decomposition–based unsupervised methods. Today it exposes Independent Component Analysis (ICA) — the go-to algorithm for blind source separation and recovering statistically independent factors. PCA, the closely related linear-decomposition method, lives in preprocessing because it is most commonly used as a feature transformer.

Key types

Usage patterns

Pattern 1: Recover independent components

use aprender::decomposition::ICA;
use aprender::primitives::Matrix;

// Suppose `mixed` is a (n_samples, n_features) matrix of mixed signals.
let mixed = Matrix::from_vec(8, 2, vec![
    1.1, 0.2, -0.8, 0.9, 0.4, -1.1, 1.5, 0.3,
    -0.7, 0.8, 1.2, -0.5, -0.3, 1.0, 0.6, -0.9,
]).expect("8x2 matrix");

let mut ica = ICA::new(2)
    .with_max_iter(200)
    .with_tolerance(1e-4)
    .with_random_state(42);
ica.fit(&mixed).expect("ica fit");

let components = ica.transform(&mixed).expect("transform");
assert_eq!(components.shape(), (8, 2));

Pattern 2: Configure ICA reproducibly for an experiment

use aprender::decomposition::ICA;

// Two reproducible runs with the same seed should produce identical mixings.
let ica_a = ICA::new(3)
    .with_random_state(123)
    .with_max_iter(500)
    .with_tolerance(1e-6);

let ica_b = ICA::new(3)
    .with_random_state(123)
    .with_max_iter(500)
    .with_tolerance(1e-6);

// Both estimators have identical hyperparameters, so given the same data
// they will converge to bit-identical un-mixing matrices.
let _ = (ica_a, ica_b);

See also

• preprocessing — PCA (linear) and TSNE (non-linear) live here
• cluster — common downstream consumer of decomposed features
• primitives — Matrix shape conventions consumed by fit / transform

Full API

Run cargo doc -p aprender-core --open for the rendered rustdoc, or browse docs.rs/aprender for the published version.