Struct ExplainabilityCallback 

pub struct ExplainabilityCallback { /* private fields */ }

Callback for computing feature attributions during training

Integrates with aprender’s interpret module to provide explainability insights during model evaluation.

Example

use entrenar::train::{ExplainabilityCallback, ExplainMethod};

let callback = ExplainabilityCallback::new(ExplainMethod::PermutationImportance)
    .with_top_k(5)
    .with_eval_samples(100);

Implementations

impl ExplainabilityCallback

pub fn new(method: ExplainMethod) -> Self

Create new explainability callback

Arguments

• method - Attribution method to use

pub fn with_top_k(self, k: usize) -> Self

Set number of top features to track

pub fn with_eval_samples(self, n: usize) -> Self

Set number of samples to use for evaluation

pub fn with_feature_names(self, names: Vec<String>) -> Self

Set feature names for interpretability

pub fn method(&self) -> ExplainMethod

Get attribution method

pub fn top_k(&self) -> usize

Get top-k setting

pub fn eval_samples(&self) -> usize

Get eval samples setting

pub fn results(&self) -> &[FeatureImportanceResult]

Get all computed results

pub fn feature_names(&self) -> Option<&[String]>

Get feature names if set

pub fn record_importances( &mut self, epoch: usize, importances: Vec<(usize, f32)>, )

Record feature importances for an epoch

Call this during on_epoch_end with computed importances

pub fn compute_permutation_importance<P>( &self, predict_fn: P, x: &[Vector<f32>], y: &[f32], ) -> Vec<(usize, f32)>

where P: Fn(&Vector<f32>) -> f32,

Compute permutation importance using aprender

Arguments

• predict_fn - Model prediction function
• x - Feature vectors
• y - Target values

pub fn compute_integrated_gradients<F>( &self, model_fn: F, sample: &Vector<f32>, baseline: &Vector<f32>, ) -> Vec<(usize, f32)>

where F: Fn(&Vector<f32>) -> f32,

Compute integrated gradients using aprender

Arguments

• model_fn - Model prediction function
• sample - Input sample to explain
• baseline - Baseline input (typically zeros)

pub fn compute_saliency<F>( &self, model_fn: F, sample: &Vector<f32>, ) -> Vec<(usize, f32)>

where F: Fn(&Vector<f32>) -> f32,

Compute saliency map using aprender

Arguments

• model_fn - Model prediction function
• sample - Input sample to explain

pub fn consistent_top_features(&self) -> Vec<(usize, f32)>

Get top features that have been consistently important across epochs

Trait Implementations

impl Debug for ExplainabilityCallback

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

Formats the value using the given formatter.

impl TrainerCallback for ExplainabilityCallback

fn on_epoch_end(&mut self, ctx: &CallbackContext) -> CallbackAction

Called after each epoch

fn name(&self) -> &'static str

Get callback name for logging

fn on_train_begin(&mut self, _ctx: &CallbackContext) -> CallbackAction

Called before training starts

fn on_train_end(&mut self, _ctx: &CallbackContext)

Called after training ends

fn on_epoch_begin(&mut self, _ctx: &CallbackContext) -> CallbackAction

Called before each epoch

fn on_step_begin(&mut self, _ctx: &CallbackContext) -> CallbackAction

Called before each training step

fn on_step_end(&mut self, _ctx: &CallbackContext) -> CallbackAction

Called after each training step

fn on_validation(&mut self, _ctx: &CallbackContext) -> CallbackAction

Called when validation is performed