cmp::covariance::ModelClusterCovariance Class Reference

Blended covariance kernel that interpolates local GP kernels using classifier probabilities. More...

#include <model_cluster.h>

Inheritance diagram for cmp::covariance::ModelClusterCovariance:

Collaboration diagram for cmp::covariance::ModelClusterCovariance:

Public Member Functions
	ModelClusterCovariance (cmp::ModelCluster *modelCluster, cmp::classifier::Classifier *classifier)
void	clearProbabilityCache () const
void	precomputeProbabilities (const Eigen::Ref< const Eigen::MatrixXd > &xObs) const
double	eval (const Eigen::VectorXd &x1, const Eigen::VectorXd &x2, const Eigen::VectorXd &par) const
double	evalGradient (const Eigen::VectorXd &x1, const Eigen::VectorXd &x2, const Eigen::VectorXd &par, const size_t &i) const
double	evalHessian (const Eigen::VectorXd &x1, const Eigen::VectorXd &x2, const Eigen::VectorXd &par, const size_t &i, const size_t &j) const
Public Member Functions inherited from cmp::covariance::Covariance
virtual	~Covariance ()=default

Static Public Member Functions
static std::shared_ptr< Covariance >	make (cmp::ModelCluster *modelCluster, cmp::classifier::Classifier *classifier)

Private Member Functions
std::string	makeProbabilityCacheKey (const Eigen::Ref< const Eigen::VectorXd > &x) const
const std::vector< double > &	getCachedProbabilities (const Eigen::Ref< const Eigen::VectorXd > &x) const

Private Attributes
cmp::ModelCluster *	pModelCluster_
	Pointer to the underlying model cluster manager.
cmp::classifier::Classifier *	pClassifier_
	Pointer to the classifier used for coordinate probability assignment.
std::unordered_map< std::string, std::vector< double > >	probabilityCache_
	Thread-local mutable cache to avoid redundant classifier evaluations.

Detailed Description

Blended covariance kernel that interpolates local GP kernels using classifier probabilities.

Mathematical Formulation The covariance function between two inputs $x$ and $x'$ is computed as:

\[ K(x, x') = \sum_{k=1}^K \sqrt{P(C=k \mid x) P(C=k \mid x')} k_k(x, x'; \theta_k) \]

where $P(C=k \mid x)$ represents the probability that input $x$ belongs to cluster $k$ (evaluated by the classifier), and $k_k$ is the kernel of the $k$-th cluster's GP.

Implementation Algorithm

1. **Cache Lookups (getCachedProbabilities)**: Generates a binary string key representing the coordinates of the input vector and queries the cache probabilityCache_ to avoid redundant classifier evaluations.
2. **Covariance Evaluation (eval)**: Fetches membership probabilities for both inputs, loops over all clusters to evaluate their respective covariance kernels, and sums the scaled products.

Constructor & Destructor Documentation

ModelClusterCovariance()

cmp::covariance::ModelClusterCovariance::ModelClusterCovariance ( cmp::ModelCluster *  modelCluster, cmp::classifier::Classifier *  classifier  )

inline

Member Function Documentation

clearProbabilityCache()

void cmp::covariance::ModelClusterCovariance::clearProbabilityCache ( ) const

inline

eval()

double cmp::covariance::ModelClusterCovariance::eval ( const Eigen::VectorXd &  x1, const Eigen::VectorXd &  x2, const Eigen::VectorXd &  par  ) const

inlinevirtual

Implements cmp::covariance::Covariance.

evalGradient()

double cmp::covariance::ModelClusterCovariance::evalGradient ( const Eigen::VectorXd &  x1, const Eigen::VectorXd &  x2, const Eigen::VectorXd &  par, const size_t &  i  ) const

inlinevirtual

Implements cmp::covariance::Covariance.

evalHessian()

double cmp::covariance::ModelClusterCovariance::evalHessian ( const Eigen::VectorXd &  x1, const Eigen::VectorXd &  x2, const Eigen::VectorXd &  par, const size_t &  i, const size_t &  j  ) const

inlinevirtual

Implements cmp::covariance::Covariance.

getCachedProbabilities()

const std::vector< double > & cmp::covariance::ModelClusterCovariance::getCachedProbabilities ( const Eigen::Ref< const Eigen::VectorXd > &  x ) const

inlineprivate

make()

static std::shared_ptr< Covariance > cmp::covariance::ModelClusterCovariance::make ( cmp::ModelCluster *  modelCluster, cmp::classifier::Classifier *  classifier  )

inlinestatic

makeProbabilityCacheKey()

std::string cmp::covariance::ModelClusterCovariance::makeProbabilityCacheKey ( const Eigen::Ref< const Eigen::VectorXd > &  x ) const

inlineprivate

precomputeProbabilities()

void cmp::covariance::ModelClusterCovariance::precomputeProbabilities ( const Eigen::Ref< const Eigen::MatrixXd > &  xObs ) const

inline

Member Data Documentation

pClassifier_

cmp::classifier::Classifier* cmp::covariance::ModelClusterCovariance::pClassifier_

private

Pointer to the classifier used for coordinate probability assignment.

pModelCluster_

cmp::ModelCluster* cmp::covariance::ModelClusterCovariance::pModelCluster_

private

Pointer to the underlying model cluster manager.

probabilityCache_

std::unordered_map<std::string, std::vector<double> > cmp::covariance::ModelClusterCovariance::probabilityCache_

mutableprivate

Thread-local mutable cache to avoid redundant classifier evaluations.

---

The documentation for this class was generated from the following file:

• include/model_cluster.h