MatchingNetworksModel<T, TInput, TOutput>

Models & Types · AiDotNet.MetaLearning.Algorithms

Matching Networks model for inference.

Implements: IModel<TInput, TOutput, ModelMetadata<T>>

How It Works

The adapted state of Matching Networks for one task: its own copy of the embedding network, the support set’s (context) embeddings and labels, and the learned matching function. Predict returns P(y | x, S) per example, [rows, NumClasses], for Tensor and Matrix outputs - a class with no support example gets probability zero, exactly as eq. 1 gives it - and the most probable class of each example for a Vector output.

Constructors

Constructor Summary
MatchingNetworksModel(IFullModel<T, TInput, TOutput> encoder, TInput supportInputs, TOutput supportLabels, MatchingNetworksOptions<T, TInput, TOutput> options, INumericOperations<T> numOps) Initializes a new instance of the MatchingNetworksModel with an untrained matching function.

Properties

Property Summary
ModelMetadata<T> Metadata { get; } Gets the model metadata.

Methods

Method Summary
ModelMetadata<T> GetModelMetadata() Gets model metadata.
TOutput Predict(TInput input) Classifies by attention over the support examples.
Vector<T> GetParameters() Gets the parameters of the model’s copy of the embedding network.
void Train(TInput inputs, TOutput targets) Training is not supported for inference models.
void UpdateParameters(Vector<T> parameters) Parameter updates are not supported for inference models.