Struct Pca 

pub struct Pca<C: PcaComponent = Unfitted> { /* private fields */ }

Principal Component Analysis, state-aware via type parameter.

Fit with Pca::fit, then project new data with Pca::transform.

transform and the basis accessors exist only on Pca<Fitted>, so projecting before fitting is a compile error rather than a runtime one:

use flow_dimensional_reduction::Pca;
let data = vec![1.0_f32, 2.0, 3.0, 4.0];
// `transform` does not exist on an unfitted model.
let _ = Pca::new(1).transform(&data, 2, 2);

The default type parameter keeps Pca::new(k) working without a turbofish.

Implementations

impl Pca<Unfitted>

pub fn new(n_components: usize) -> Self

Create an unfitted PCA requesting n_components components.

The actual count is clamped to the feature count d during Pca::fit.

pub fn fit(self, data: &[f32], n: usize, d: usize) -> PcaResult<Pca<Fitted>>

Fit to n × d row-major data, consuming the unfitted model.

Means and the covariance matrix are accumulated in f64 regardless of the f32 input, then downcast — n can be large enough that f32 accumulation loses significant precision.

Errors

PcaError::EmptyData if n == 0 or d == 0; PcaError::InsufficientData if n < 2; PcaError::DimensionMismatch if data.len() != n * d; PcaError::SvdFailed if the decomposition fails.

impl Pca<Fitted>

pub fn transform(&self, data: &[f32], n: usize, d: usize) -> PcaResult<Vec<f32>>

Project n × d row-major data onto the fitted axes.

Returns n × n_components row-major.

Errors

PcaError::DimensionMismatch if data.len() != n * d; PcaError::FeatureMismatch if d differs from the fitted feature count.

pub fn components(&self) -> &[f32]

Principal axes, k * d row-major: axis i occupies [i*d .. (i+1)*d].

pub fn components_shape(&self) -> (usize, usize)

Shape of Self::components as (k, d).

pub fn explained_variance_ratio(&self) -> &[f32]

Fraction of total variance per component, descending.

pub fn mean(&self) -> &[f32]

Column means of the training data.

impl<C: PcaComponent> Pca<C>

Available in every state — delegates to the trait method, since a generic C cannot be pattern-matched.

pub fn n_components(&self) -> usize

Component count: requested before fitting, actual (clamped) after.

Trait Implementations

impl<C: Clone + PcaComponent> Clone for Pca<C>

fn clone(&self) -> Pca<C>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<C: Debug + PcaComponent> Debug for Pca<C>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.