NeuralProcessModel<T, TInput, TOutput>

Models & Types · AiDotNet.MetaLearning.Algorithms

Adapted model for Neural Process family algorithms.

Inheritance: ModelWrapperBase<T, TInput, TOutput> → MetaLearningModelBase<T, TInput, TOutput> → NeuralProcessModel<T, TInput, TOutput>

Implements: IAdaptedMetaModel<T>

For Beginners

This is the output of a Neural Process after it has seen some context examples. It stores the aggregated representation of what the model learned from those examples and uses it to make predictions at new target points. Unlike gradient-based approaches, adaptation happens in a single forward pass.

Example

var baseModel = new SimpleRegression<float>();
var parameters = new Vector<float>((int)baseModel.ParameterCount);
            
// the context examples, encoded and aggregated into one representation
var aggregatedRepresentation = new Vector<float>(new float[] { 0.2f, -0.1f, 0.4f });
            
var targetInputs = new Matrix<float>(new float[,] { { 1.0f, 2.0f }, { 3.0f, 4.0f } });
var targetOutputs = new Vector<float>(new float[] { 3.1f, 7.2f });
            
var result = new AiModelBuilder<float, Matrix<float>, Vector<float>>()
    .ConfigureModel(new NeuralProcessModel<float, Matrix<float>, Vector<float>>(
        baseModel, parameters, aggregatedRepresentation))
    .Build(targetInputs, targetOutputs);
            
Vector<float> prediction = result.Predict(targetInputs);

Constructors

Constructor Summary
NeuralProcessModel(IFullModel<T, TInput, TOutput> model, Vector<T> parameters, Vector<T> contextRepresentation) Initializes the adapted model from the meta-model and the parameters conditioned on the context set.

Properties

Property Summary
Vector<T> AdaptedSupportFeatures { get; }
Vector<T> ParameterModulationFactors { get; }

Methods

Method Summary
IFullModel<T, TInput, TOutput> WithParameters(Vector<T> parameters)
TOutput Predict(TInput input)