Struct MagnitudeImportance 

pub struct MagnitudeImportance { /* private fields */ }

Magnitude-based importance estimator.

The simplest importance metric: larger weights are more important. This is the original method from Han et al. (2015).

Formula

• L1: importance = |w|
• L2: importance = w²

Advantages

• No calibration data required
• Very fast to compute
• Good baseline for comparison

Limitations

• Ignores activation patterns
• Can underestimate importance of small but critical weights

Implementations

impl MagnitudeImportance

pub fn l1() -> MagnitudeImportance

Create L1 magnitude importance estimator.

Formula

importance = |w|

pub fn l2() -> MagnitudeImportance

Create L2 magnitude importance estimator.

Formula

importance = w²

pub fn with_norm(norm: NormType) -> MagnitudeImportance

Create magnitude importance with specified norm.

pub fn norm(&self) -> NormType

Get the norm type.

pub fn compute_for_weights( &self, weights: &Tensor, ) -> Result<Tensor, PruningError>

Compute importance for a single weight tensor.

Arguments

• weights - Weight tensor to compute importance for

Returns

Importance scores with same shape as weights.

Toyota Way: Jidoka

Validates for NaN/Inf values and returns error immediately.

Trait Implementations

impl Clone for MagnitudeImportance

fn clone(&self) -> MagnitudeImportance

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MagnitudeImportance

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

Formats the value using the given formatter.

impl Importance for MagnitudeImportance

fn compute( &self, module: &dyn Module, _context: Option<&CalibrationContext>, ) -> Result<ImportanceScores, PruningError>

Compute importance scores for a module’s parameters.

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

Name of this importance method.

fn requires_calibration(&self) -> bool

Whether this method requires calibration data.