Struct Dataset 

#[non_exhaustive]
pub struct Dataset {
    pub features: Vec<Vec<f64>>,
    pub target: Vec<f64>,
    pub feature_names: Vec<String>,
    pub target_name: String,
    pub class_labels: Option<Vec<String>>,
    /* private fields */
}

A tabular dataset with features and a target column.

Features are stored column-major (features[feature_idx][sample_idx]) for cache-friendly access during tree split evaluation.

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

features: Vec<Vec<f64>>

Feature columns: features[feature_idx][sample_idx].

target: Vec<f64>

Target values: target[sample_idx].

feature_names: Vec<String>

Feature column names.

target_name: String

Target column name.

class_labels: Option<Vec<String>>

Class label mapping (index → label string) for classification tasks.

Implementations

impl Dataset

pub fn new( features: Vec<Vec<f64>>, target: Vec<f64>, feature_names: Vec<String>, target_name: impl Into<String>, ) -> Self

Create a dataset from pre-computed features and target.

Panics

Panics if feature columns have mismatched lengths, or if feature_names.len() != features.len().

pub fn from_matrix( matrix: DenseMatrix, target: Vec<f64>, feature_names: Vec<String>, target_name: impl Into<String>, ) -> Self

Create a dataset from a DenseMatrix, target, and column names.

The features field is populated from the matrix for backward compat.

pub fn matrix(&self) -> &DenseMatrix

The contiguous column-major feature matrix.

Lazily built from features on first access. Subsequent calls return the cached matrix without recomputation.

pub fn n_samples(&self) -> usize

Number of samples (rows).

pub fn n_features(&self) -> usize

Number of features (columns).

pub fn n_classes(&self) -> usize

Number of unique classes in the target (for classification).

pub fn feature(&self, idx: usize) -> &[f64]

Get a single feature column by index.

pub fn sample(&self, idx: usize) -> Vec<f64>

Get a single sample (row) as a vector of feature values.

pub fn feature_matrix(&self) -> Vec<Vec<f64>>

Get the feature matrix as row-major [n_samples][n_features].

pub fn flat_feature_matrix(&mut self) -> &[f64]

Get a contiguous row-major feature buffer, computing on first call.

Layout: [sample_0_feat_0, sample_0_feat_1, ..., sample_n_feat_m]. Subsequent calls return the cached slice without recomputation.

pub fn sample_row<'a>(&self, cache: &'a [f64], idx: usize) -> &'a [f64]

Get a zero-copy row slice from a pre-computed flat feature buffer.

cache should be the result of Dataset::flat_feature_matrix.

pub fn subset(&self, indices: &[usize]) -> Self

Create a subset of this dataset with the given sample indices.

pub fn sync_matrix(&mut self)

Clear the cached matrix so it will be lazily rebuilt from features on the next call to matrix().

Call this after mutating features in place (e.g. after a transformer’s transform() step).

pub fn invalidate_matrix(&mut self)

Mark the matrix cache as stale after in-place feature mutations.

The matrix will be lazily rebuilt from features on next access.

pub fn validate_finite(&self) -> Result<()>

Returns Err(InvalidData) if any feature or target value is NaN or ±Inf.

pub fn validate_no_inf(&self) -> Result<()>

Returns Err(InvalidData) if any feature or target value is ±Inf.

Unlike validate_finite, this allows NaN values (useful for imputers that intentionally handle NaN).

pub fn with_class_labels(self, labels: Vec<String>) -> Self

Attach class labels for classification.

pub fn from_sparse( csc: CscMatrix, target: Vec<f64>, feature_names: Vec<String>, target_name: impl Into<String>, ) -> Self

Create a dataset from a sparse CSC matrix.

The features field is left empty. Call ensure_dense before accessing features directly on a sparse dataset.

pub fn is_sparse(&self) -> bool

Whether this dataset uses sparse storage.

pub fn sparse_csc(&self) -> Option<&CscMatrix>

Get the sparse CSC matrix if available.

pub fn sparse_csr(&self) -> Option<CsrMatrix>

Get the sparse CSR matrix (converted from CSC on demand).

pub fn summary(&self) -> Vec<ColumnStats>

Compute descriptive statistics for every feature column and the target.

Returns one ColumnStats per feature (in order) followed by one for the target column. NaN values are excluded from all computations. Standard deviation uses ddof=1 (sample std) to match pandas.

pub fn describe(&self)

Print a pandas-style descriptive statistics table to stdout.

Internally calls summary().

pub fn ensure_dense(&mut self)

Populate the features field from sparse storage.

No-op if the dataset is already dense. After calling this, features[j][i] is available as usual.

Trait Implementations

impl Clone for Dataset

fn clone(&self) -> Dataset

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Dataset

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

Formats the value using the given formatter.