MultiTaskGaussianProcess<T>
Models & Types · AiDotNet.GaussianProcesses
Implements a Multi-Task Gaussian Process for modeling multiple correlated outputs.
Inheritance: ModelBase<T, Matrix<T>, Vector<T>> → GaussianProcessBase<T> → MultiTaskGaussianProcess<T>
Implements: IGeneratedParameterRegistrar<T>
For Beginners
A Multi-Task GP models multiple related output variables simultaneously, learning the correlations between tasks to improve predictions for all of them.
Example scenarios:
- Predicting temperature, humidity, and pressure at weather stations (related measurements)
- Forecasting sales across multiple product lines (correlated markets)
- Modeling grades across subjects for students (abilities correlate)
Why use Multi-Task GP instead of separate GPs?
- Information sharing: If task A has lots of data and task B has little,
the Multi-Task GP can use A’s data to help predict B
- Correlation modeling: Learns how tasks relate (e.g., when temperature rises,
ice cream sales increase)
- Better uncertainty: More accurate confidence intervals by considering
task relationships
The model uses a coregionalization approach:
- A base kernel captures input similarity
- A task correlation matrix captures how tasks relate
- The combined kernel is their product (Kronecker structure)
Constructors
| Constructor | Summary |
|---|---|
MultiTaskGaussianProcess(IKernelFunction<T> kernel, int numTasks, double? noiseVariance, bool learnTaskCorrelations, MatrixDecompositionType decompositionType, int hyperparameterOptimizationSteps) |
Initializes a new Multi-Task Gaussian Process. |
Methods
| Method | Summary |
|---|---|
Matrix<T> GetTaskCorrelations() |
Gets the learned task correlation matrix. |
ValueTuple<T, T> Predict(Vector<T> x) |
IFullModel compliance: Predict single point returning first task’s mean. |
ValueTuple<Vector<T>, Vector<T>> PredictMultiTask(Vector<T> x) |
Predicts all task outputs for a new input point. |
void Fit(Matrix<T> X, Matrix<T> Y) |
IFullModel compliance: Fit with single-output vector (uses first task only). |
void Fit(Matrix<T> X, Vector<T> y) |
IFullModel compliance: Fit with single-output vector (uses first task only). |
void UpdateKernel(IKernelFunction<T> kernel) |
Updates the kernel for this multi-task GP. |