Table of Contents

Class WeightedSampler<T>

Namespace
AiDotNet.Data.Sampling
Assembly
AiDotNet.dll

A sampler that samples indices based on their weights.

public class WeightedSampler<T> : WeightedSamplerBase<T>, IWeightedSampler<T>, IDataSampler

Type Parameters

T

The numeric type for weights.

Inheritance
WeightedSampler<T>
Implements
Inherited Members

Remarks

WeightedSampler selects samples with probability proportional to their weights. This is useful for handling class imbalance, importance sampling, or focusing training on harder examples.

For Beginners: This sampler lets you control how often each sample appears:

  • Higher weight = more likely to be selected
  • Lower weight = less likely to be selected

Common uses:

  • Class imbalance: Give higher weights to underrepresented classes
  • Hard example mining: Give higher weights to samples the model struggles with
  • Curriculum learning: Adjust weights during training based on difficulty

Example:

// Dataset has 900 cats, 100 dogs
// Give dogs 9x higher weight to balance
var weights = labels.Select(l => l == "dog" ? 9.0f : 1.0f).ToList();
var sampler = new WeightedSampler<float>(weights);

Constructors

WeightedSampler(IEnumerable<T>, int?, bool, int?)

Initializes a new instance of the WeightedSampler class.

public WeightedSampler(IEnumerable<T> weights, int? numSamples = null, bool replacement = true, int? seed = null)

Parameters

weights IEnumerable<T>

The weight for each sample. Must be non-negative.

numSamples int?

Number of samples to draw per epoch. Defaults to dataset size.

replacement bool

Whether to sample with replacement.

seed int?

Optional random seed for reproducibility.

Exceptions

ArgumentNullException

Thrown when weights is null.

ArgumentException

Thrown when weights is empty or contains negative values.

Methods

CreateBalancedWeights(IReadOnlyList<int>, int)

Creates weights that balance class frequencies.

public static T[] CreateBalancedWeights(IReadOnlyList<int> labels, int numClasses)

Parameters

labels IReadOnlyList<int>

The class labels for each sample.

numClasses int

The number of classes.

Returns

T[]

Weights that inversely weight by class frequency.

Remarks

This helper method creates weights such that each class has equal total weight, effectively balancing an imbalanced dataset.

For Beginners: If you have 900 cats and 100 dogs, calling this method will give dogs 9x the weight of cats, so they get sampled equally often.

GetIndicesCore()

Core implementation for generating indices. Override this in derived classes.

protected override IEnumerable<int> GetIndicesCore()

Returns

IEnumerable<int>

An enumerable of sample indices.

Edit this page