Struct ImportanceSampler 

pub struct ImportanceSampler { /* private fields */ }

Importance sampling for biased data handling

This sampler adjusts sampling probabilities to correct for dataset bias or to emphasize certain types of samples during training. It allows for sophisticated control over the sampling distribution through importance weights.

Examples

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

// Create importance weights (higher = more important)
let importance_weights = vec![0.1, 0.5, 1.0, 0.3, 0.8];
let sampler = ImportanceSampler::new(importance_weights, 3, true)
    .with_temperature(0.5)
    .with_generator(42);

let indices: Vec<usize> = sampler.iter().collect();
assert_eq!(indices.len(), 3);

Implementations

impl ImportanceSampler

pub fn new( importance_weights: Vec<f64>, num_samples: usize, replacement: bool, ) -> Self

Create a new importance sampler

Arguments

• importance_weights - Importance weights for each sample (higher = more important)
• num_samples - Number of samples to select
• replacement - Whether to sample with replacement

Panics

Panics if importance_weights is empty or contains negative values.

Examples

use torsh_data::sampler::ImportanceSampler;

let weights = vec![1.0, 2.0, 0.5, 3.0]; // Index 3 is most important
let sampler = ImportanceSampler::new(weights, 2, false);

pub fn with_temperature(self, temperature: f64) -> Self

Set temperature for softmax scaling of importance weights

Temperature controls the sharpness of the importance distribution:

• Higher temperature (> 1.0) = more uniform sampling
• Lower temperature (< 1.0) = more biased toward high importance samples
• Temperature = 1.0 = no scaling

Arguments

• temperature - Temperature value (must be positive)

Examples

use torsh_data::sampler::ImportanceSampler;

let weights = vec![1.0, 2.0, 3.0];
let sampler = ImportanceSampler::new(weights, 2, true)
    .with_temperature(0.5); // More emphasis on high importance

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

Set random generator seed

Arguments

• seed - Random seed for reproducible sampling

pub fn importance_weights(&self) -> &[f64]

Get the importance weights

pub fn num_samples(&self) -> usize

Get the number of samples

pub fn replacement(&self) -> bool

Check if sampling with replacement

pub fn temperature(&self) -> f64

Get the temperature

pub fn generator(&self) -> Option<u64>

Get the generator seed if set

pub fn update_weights(&mut self, new_weights: Vec<f64>)

Update importance weights

Arguments

• new_weights - New importance weights

Panics

Panics if new_weights has different length than original weights.

pub fn sampling_stats(&self) -> ImportanceStats

Get sampling statistics

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