Class ProtoNetsOptions<T, TInput, TOutput>

Namespace

AiDotNet.MetaLearning.Options

Assembly

AiDotNet.dll

Configuration options for Prototypical Networks (ProtoNets) algorithm.

public class ProtoNetsOptions<T, TInput, TOutput> : IMetaLearnerOptions<T>

Type Parameters

T

The numeric data type used for calculations (e.g., float, double).

TInput

The input data type (e.g., Matrix<T>, Tensor<T>).

TOutput

The output data type (e.g., Vector<T>, Tensor<T>).

Inheritance

object

ProtoNetsOptions<T, TInput, TOutput>

Implements

IMetaLearnerOptions<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

Prototypical Networks learn a metric space where classification is performed by computing distances to prototype representations of each class. Each prototype is the mean vector of the support set examples for that class in the learned embedding space.

For Beginners: ProtoNets is one of the simplest and most effective few-shot learning methods:

1. Use a neural network to convert images/data into feature vectors
2. For each class in a task, compute the "prototype" (average feature vector)
3. To classify a new example, find the nearest prototype
4. Train the network to make same-class examples cluster together

Unlike MAML, ProtoNets doesn't need gradient updates at test time - just compute prototypes and measure distances!

Constructors

ProtoNetsOptions(IFullModel<T, TInput, TOutput>)

Initializes a new instance of the ProtoNetsOptions class with the required meta-model.

public ProtoNetsOptions(IFullModel<T, TInput, TOutput> metaModel)

Parameters

metaModel IFullModel<T, TInput, TOutput>

The feature encoder to be trained (required).

Examples

// Create ProtoNets with minimal configuration
var options = new ProtoNetsOptions<double, Tensor<double>, Tensor<double>>(myEncoder);
var protoNets = new ProtoNetsAlgorithm<double, Tensor<double>, Tensor<double>>(options);

// Create ProtoNets with custom configuration
var options = new ProtoNetsOptions<double, Tensor<double>, Tensor<double>>(myEncoder)
{
    DistanceFunction = ProtoNetsDistanceFunction.Cosine,
    NormalizeFeatures = true,
    Temperature = 0.5
};

Exceptions

ArgumentNullException

Thrown when metaModel is null.

Properties

AdaptationSteps

Gets or sets the number of adaptation steps.

public int AdaptationSteps { get; set; }

Property Value

int

Default is 1 (ProtoNets uses non-parametric adaptation).

Remarks

ProtoNets doesn't use gradient-based adaptation steps like MAML. This value is kept for interface compatibility. The "adaptation" in ProtoNets is simply computing class prototypes, which is done in one step.

CheckpointFrequency

Gets or sets how often to save checkpoints.

public int CheckpointFrequency { get; set; }

Property Value

int

DataLoader

Gets or sets the episodic data loader for sampling tasks. Default: null (tasks must be provided manually to MetaTrain).

public IEpisodicDataLoader<T, TInput, TOutput>? DataLoader { get; set; }

Property Value

IEpisodicDataLoader<T, TInput, TOutput>

DistanceFunction

Gets or sets the distance function for measuring similarity between embeddings.

public ProtoNetsDistanceFunction DistanceFunction { get; set; }

Property Value

ProtoNetsDistanceFunction

The distance function. Default is Euclidean.

Remarks

The distance function determines how similarity is measured between query embeddings and class prototypes. Euclidean distance is the default and works well for most applications.

For Beginners: Start with Euclidean. Try Cosine if your features are normalized. Use Mahalanobis if you have domain knowledge about feature correlations.

EnableCheckpointing

Gets or sets whether to save model checkpoints.

public bool EnableCheckpointing { get; set; }

Property Value

bool

EvaluationFrequency

Gets or sets how often to evaluate the meta-learner.

public int EvaluationFrequency { get; set; }

Property Value

int

EvaluationTasks

Gets or sets the number of tasks to use for evaluation.

public int EvaluationTasks { get; set; }

Property Value

int

GradientClipThreshold

Gets or sets the maximum gradient norm for gradient clipping.

public double? GradientClipThreshold { get; set; }

Property Value

double?

The gradient clip threshold, or null to disable. Default is 10.0.

InnerLearningRate

Gets or sets the learning rate for the inner loop (task adaptation).

public double InnerLearningRate { get; set; }

Property Value

double

The inner learning rate. Default is 0.01.

Remarks

Note: ProtoNets doesn't perform gradient-based inner loop adaptation like MAML. This value is kept for interface compatibility but is not used in the core algorithm. Prototype computation is non-parametric.

LossFunction

Gets or sets the loss function for training. Default: null (uses cross-entropy loss internally).

public ILossFunction<T>? LossFunction { get; set; }

Property Value

ILossFunction<T>

Remarks

ProtoNets typically uses cross-entropy loss with softmax over negative distances. This encourages the correct class prototype to be closer than other prototypes.

MahalanobisScaling

Gets or sets the scaling factor for Mahalanobis distance.

public double MahalanobisScaling { get; set; }

Property Value

double

The Mahalanobis scaling factor. Default is 1.0.

Remarks

This is a simplified scaling factor used when DistanceFunction is Mahalanobis. In a full implementation, this would be replaced by a learned covariance matrix.

MetaBatchSize

Gets or sets the number of tasks to sample per meta-training iteration.

public int MetaBatchSize { get; set; }

Property Value

int

The meta-batch size. Default is 4.

Remarks

Each training step samples this many tasks (episodes), computes loss on each, and averages the gradients before updating the encoder.

MetaModel

Gets or sets the meta-model (feature encoder) to be trained. This is the only required property.

public IFullModel<T, TInput, TOutput> MetaModel { get; set; }

Property Value

IFullModel<T, TInput, TOutput>

Remarks

The meta-model should be a feature encoder that maps inputs to an embedding space where same-class examples are close together and different-class examples are far apart.

For Beginners: This is typically a CNN for images or an MLP for tabular data. The network learns to output feature vectors that cluster by class.

MetaOptimizer

Gets or sets the optimizer for feature encoder updates. Default: null (uses built-in Adam optimizer with OuterLearningRate).

public IGradientBasedOptimizer<T, TInput, TOutput>? MetaOptimizer { get; set; }

Property Value

IGradientBasedOptimizer<T, TInput, TOutput>

NormalizeFeatures

Gets or sets whether to L2-normalize feature embeddings.

public bool NormalizeFeatures { get; set; }

Property Value

bool

True to normalize features; false otherwise. Default is false.

Remarks

When enabled, feature vectors are normalized to unit length before computing prototypes and distances. This can improve stability and is recommended when using cosine distance.

For Beginners: Enable this if you're using Cosine distance, or if you notice that feature magnitudes vary widely.

NumMetaIterations

Gets or sets the total number of meta-training iterations.

public int NumMetaIterations { get; set; }

Property Value

int

The number of meta-iterations. Default is 1000.

OuterLearningRate

Gets or sets the learning rate for the outer loop (encoder training).

public double OuterLearningRate { get; set; }

Property Value

double

The outer learning rate. Default is 0.001.

Remarks

This controls how quickly the feature encoder is updated during training. A typical range is 0.0001 to 0.01.

For Beginners: This is how fast the embedding network learns. Start with 0.001 and adjust based on training stability.

RandomSeed

Gets or sets the random seed for reproducibility.

public int? RandomSeed { get; set; }

Property Value

int?

Temperature

Gets or sets the temperature for softmax scaling.

public double Temperature { get; set; }

Property Value

double

The temperature value. Default is 1.0.

Remarks

Temperature controls the sharpness of the probability distribution: - Lower temperature (< 1.0): Sharper, more confident predictions - Higher temperature (> 1.0): Softer, more uniform predictions

For Beginners: Leave at 1.0 unless you want to calibrate confidence. Lower values make the model more "certain" about its predictions.

UseAdaptiveClassScaling

Gets or sets whether to use adaptive class-specific scaling factors.

public bool UseAdaptiveClassScaling { get; set; }

Property Value

bool

True to use adaptive scaling; false otherwise. Default is false.

Remarks

When enabled, learns per-class scaling factors for distances. This allows the model to handle classes with different intra-class variances.

UseAttentionMechanism

Gets or sets whether to use an attention mechanism for prototype computation.

public bool UseAttentionMechanism { get; set; }

Property Value

bool

True to use attention; false for simple averaging. Default is false.

Remarks

When enabled, uses learned attention weights to compute weighted prototypes instead of simple averaging. This can help focus on more informative examples.

For Beginners: Start with false (simple averaging). Enable if you have noisy or heterogeneous support sets.

UseFirstOrder

Gets or sets whether to use first-order approximation.

public bool UseFirstOrder { get; set; }

Property Value

bool

Default is true since ProtoNets doesn't use gradient-based inner loop.

Remarks

For ProtoNets, this option is less relevant because adaptation is non-parametric (prototype computation, not gradient descent). This property is provided for interface compatibility with IMetaLearnerOptions.

Methods

Clone()

Creates a deep copy of the ProtoNets options.

public IMetaLearnerOptions<T> Clone()

Returns

IMetaLearnerOptions<T>

A new ProtoNetsOptions instance with the same configuration.

IsValid()

Validates that all ProtoNets configuration options are properly set.

public bool IsValid()

Returns

bool

True if the configuration is valid; otherwise, false.

Remarks

Checks: - MetaModel is set - Learning rate is positive - Temperature is positive - Batch sizes and iteration counts are positive