Struct ImportanceSampler 

pub struct ImportanceSampler { /* private fields */ }

Importance sampler for adaptive sample selection.

This sampler selects samples based on importance scores, which can represent various metrics like loss values, prediction confidence, or gradient norms. High-importance samples are selected more frequently, making this ideal for active learning and hard negative mining.

Applications

• Active Learning: Sample uncertain or informative examples
• Hard Negative Mining: Focus on difficult examples
• Curriculum Learning: Gradually increase sample difficulty
• Online Learning: Adapt to changing data distributions

Implementations

impl ImportanceSampler

pub fn new(importance_scores: Vec<f32>) -> Self

Create a new importance sampler.

Arguments

• importance_scores - Vector of importance values for each sample

Examples

use torsh_data::sampler::{Sampler, ImportanceSampler};

// Higher scores = more important
let scores = vec![0.1, 0.8, 0.3, 0.9, 0.2];
let sampler = ImportanceSampler::new(scores);

// Samples 1 and 3 will be selected more frequently

pub fn with_temperature(self, temperature: f32) -> Self

Set the temperature for importance sampling.

Higher temperature makes sampling more uniform, lower temperature makes it more focused on high-importance samples.

Arguments

• temperature - Temperature parameter (> 0.0)

Examples

use torsh_data::sampler::ImportanceSampler;

let scores = vec![0.1, 0.8, 0.3];
let sampler = ImportanceSampler::new(scores)
    .with_temperature(2.0); // More uniform sampling

pub fn with_adaptive(self, adaptive: bool, update_rate: f32) -> Self

Enable adaptive importance updates.

When enabled, importance scores can be updated based on recent sampling feedback to adapt to changing data characteristics.

Arguments

• adaptive - Whether to enable adaptive updates
• update_rate - Rate of adaptation (0.0 to 1.0)

pub fn with_generator(self, seed: u64) -> Self

Set random generator seed.

pub fn importance_scores(&self) -> &[f32]

Get the importance scores.

pub fn temperature(&self) -> f32

Get the temperature parameter.

pub fn is_adaptive(&self) -> bool

Check if adaptive updates are enabled.

pub fn update_importance_scores(&mut self, new_scores: Vec<f32>)

Update importance scores (for adaptive sampling).

Arguments

• new_scores - Updated importance scores

Examples

use torsh_data::sampler::ImportanceSampler;

let mut sampler = ImportanceSampler::new(vec![0.1, 0.5, 0.3])
    .with_adaptive(true, 0.1);

// Update based on new loss values
let new_losses = vec![0.2, 0.8, 0.1];
sampler.update_importance_scores(new_losses);

Trait Implementations

impl Clone for ImportanceSampler

fn clone(&self) -> ImportanceSampler

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ImportanceSampler

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

Formats the value using the given formatter.

impl Sampler for ImportanceSampler

type Iter = SamplerIterator

Iterator type returned by the sampler

fn iter(&self) -> Self::Iter

Create an iterator over indices

fn len(&self) -> usize

Total number of samples that will be yielded

fn is_empty(&self) -> bool

Check if sampler is empty

fn into_batch_sampler( self, batch_size: usize, drop_last: bool, ) -> BatchingSampler<Self>

where Self: Sized,

Convert this sampler into a batch sampler

fn into_distributed( self, num_replicas: usize, rank: usize, ) -> DistributedWrapper<Self>

where Self: Sized,

Create a distributed version of this sampler