Struct ConfidentLearning 

pub struct ConfidentLearning { /* private fields */ }

Confident Learning for label noise detection.

Identifies mislabeled samples by finding confident joint distribution discrepancies between noisy labels and model predictions.

Reference

• Northcutt, C., et al. (2021). Confident Learning: Estimating Uncertainty in Dataset Labels. JAIR.

Implementations

impl ConfidentLearning

pub fn new(n_classes: usize) -> Self

Create new confident learning instance.

pub fn with_threshold(n_classes: usize, threshold: f32) -> Self

Create with custom threshold.

pub fn find_label_issues( &self, labels: &[usize], pred_probs: &[Vector<f32>], ) -> Vec<usize>

Find potentially mislabeled samples.

Arguments

• labels - Noisy labels
• pred_probs - Model-predicted class probabilities

Returns

Indices of samples likely to be mislabeled.

pub fn compute_confident_joint( &self, labels: &[usize], pred_probs: &[Vector<f32>], ) -> Vec<Vec<f32>>

Compute confident joint matrix.

Element (i,j) represents count of samples with noisy label i and confident predicted label j.

pub fn estimate_noise_matrix( &self, labels: &[usize], pred_probs: &[Vector<f32>], ) -> Vec<Vec<f32>>

Estimate noise transition matrix P(noisy=i | true=j).

Trait Implementations

impl Clone for ConfidentLearning

fn clone(&self) -> ConfidentLearning

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ConfidentLearning

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

Formats the value using the given formatter.