Struct IntegratedGradients 

pub struct IntegratedGradients { /* private fields */ }

Integrated Gradients

Integrated Gradients is a method that attributes the prediction of a model to its input features by integrating gradients along a path from a baseline to the input.

Reference: Sundararajan et al., “Axiomatic Attribution for Deep Networks”, ICML 2017

Implementations

impl IntegratedGradients

pub fn new( baseline_type: BaselineType, num_steps: usize, device: Arc<dyn Device>, ) -> Self

Create a new integrated gradients explainer

pub fn generate( &self, model: &dyn Module, input: &Tensor, target_class: usize, ) -> Result<Tensor>

Generate integrated gradients attribution map

Arguments

• model - The model to explain
• input - Input image tensor [C, H, W] or [N, C, H, W]
• target_class - Class index to generate attribution for

Returns

Attribution map showing feature importance