Struct FeatureImportance 

pub struct FeatureImportance { /* private fields */ }

Tracks per-feature importance via accumulated split gains.

Each time a tree splits on a feature, the gain from that split is added to the feature’s running total. This produces a simple but effective measure of feature importance for gradient boosted trees.

Example

use irithyll::metrics::FeatureImportance;

let mut fi = FeatureImportance::new(3);
fi.update(0, 1.5);
fi.update(2, 0.8);
fi.update(0, 0.3);

let top = fi.top_k(2);
assert_eq!(top[0].0, 0); // feature 0 has highest gain

Implementations

impl FeatureImportance

pub fn new(n_features: usize) -> Self

Create a new tracker for n_features features, all initialized to zero.

pub fn update(&mut self, feature_idx: usize, gain: f64)

Accumulate gain for the feature at feature_idx.

Panics

Panics if feature_idx >= n_features.

pub fn importances(&self) -> &[f64]

Raw accumulated gains for all features.

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

Importances normalized to sum to 1.0.

Returns a vector of zeros if total accumulated gain is zero.

pub fn top_k(&self, k: usize) -> Vec<(usize, f64)>

Top-k features by accumulated gain, sorted descending.

Returns (feature_index, gain) pairs. If k > n_features, all features are returned.

pub fn n_features(&self) -> usize

Number of features being tracked.

pub fn total_gain(&self) -> f64

Total accumulated gain across all features.

pub fn reset(&mut self)

Reset all feature importances to zero.

Trait Implementations

impl Clone for FeatureImportance

fn clone(&self) -> FeatureImportance

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for FeatureImportance

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

Formats the value using the given formatter.