Struct SensitivityAnalyzer 

pub struct SensitivityAnalyzer { /* private fields */ }

Measures feature importance via finite-difference sensitivity analysis.

Implementations

impl SensitivityAnalyzer

pub fn new(input_dim: usize) -> Self

Create an analyzer for inputs of size input_dim.

pub fn input_sensitivity<F>( &self, input: &Array1<f32>, forward_fn: F, eps: f32, ) -> ModelResult<Array1<f32>>

where F: Fn(&Array1<f32>) -> ModelResult<Array1<f32>>,

Estimate per-feature sensitivity using finite differences.

For each feature i, computes ||f(x + eps*e_i) - f(x)|| / eps.

Arguments

• input — base input vector (length must equal input_dim)
• forward_fn — model forward pass (called input_dim + 1 times)
• eps — perturbation magnitude (default 1e-3 is reasonable)

pub fn rank_features<F>( &self, inputs: &[Array1<f32>], forward_fn: F, eps: f32, ) -> ModelResult<Vec<(usize, f32)>>

where F: Fn(&Array1<f32>) -> ModelResult<Array1<f32>>,

Rank features by average sensitivity across multiple input vectors.

Returns a sorted list of (feature_index, avg_sensitivity) in descending order.