Abstract base class for Bayesian nonparametric processes. More...

#include <Process.hpp>

Inheritance diagram for Process:

Public Member Functions
	Process (Data &d, Params &p)
	Constructor initializing process with data and parameters.

virtual double	gibbs_prior_existing_cluster (int cls_idx, int obs_idx) const =0
	Compute prior probability for assigning observation to existing cluster.

virtual Eigen::VectorXd	gibbs_prior_existing_clusters (int obs_idx) const =0
	Compute prior probabilities for assigning observation to all existing clusters.

virtual double	gibbs_prior_new_cluster () const =0
	Compute prior probability for creating a new cluster.

virtual double	prior_ratio_split (int ci, int cj) const =0
	Compute prior ratio for split move in split-merge algorithm.

virtual double	prior_ratio_merge (int size_old_ci, int size_old_cj) const =0
	Compute prior ratio for merge move in split-merge algorithm.

virtual double	prior_ratio_shuffle (int size_old_ci, int size_old_cj, int ci, int cj) const =0
	Compute prior ratio for shuffle move in split-merge algorithm.

void	set_old_allocations (const Eigen::VectorXi &new_allocations)
	Store current allocations for potential rollback.

void	set_idx_i (int i)
	Set index of first observation in split-merge pair.

void	set_idx_j (int j)
	Set index of second observation in split-merge pair.

virtual void	update_params ()=0
	Update process parameters during MCMC sampling.

virtual	~Process ()
	Virtual destructor for proper cleanup of derived classes.

Protected Attributes
Data &	data
	Reference to the data object containing observations and allocations.

Params &	params
	Reference to the parameters object containing process hyperparameters.

Eigen::VectorXi	old_allocations
	Storage for previous allocations to enable rollback in case of rejection.

int	idx_i
	Index of first observation involved in split-merge move.

int	idx_j
	Index of second observation involved in split-merge move.

const double	log_a = log(params.a)
	Precomputed logarithm of total mass parameter for efficiency.

Detailed Description

Abstract base class for Bayesian nonparametric processes.

This class provides the foundation for implementing different types of Bayesian nonparametric processes including Dirichlet Process (DP), Normalized Generalized Gamma Process (NGGP), and their weighted variants (DPW, NGGPW). It defines the interface for prior computations needed in both Gibbs sampling and split-merge MCMC algorithms.

The class handles:

Prior probability computations for existing and new clusters
Prior ratio calculations for split-merge moves
Parameter updates during MCMC sampling
State management for split-merge algorithms

Derived classes must implement the pure virtual methods to define their specific process characteristics (DP, NGGP, NGGPW, DPW).

See also: DP, NGGP, NGGPW, DPW

Constructor & Destructor Documentation

◆ Process()

Process::Process	(	Data &	d,
		Params &	p )

inline

Constructor initializing process with data and parameters.

Parameters

d	Reference to Data object containing observations and current allocations
p	Reference to Params object containing process hyperparameters

Initializes the process and stores a copy of current allocations for potential rollback operations during split-merge moves.

◆ ~Process()

virtual Process::~Process ( )

inlinevirtual

Virtual destructor for proper cleanup of derived classes.

Member Function Documentation

◆ gibbs_prior_existing_cluster()

virtual double Process::gibbs_prior_existing_cluster	(	int	cls_idx,
		int	obs_idx ) const

pure virtual

Compute prior probability for assigning observation to existing cluster.

Parameters

cls_idx	Index of the existing cluster
obs_idx	Index of the observation to be assigned

Returns: Log prior probability of assignment to existing cluster

This method computes the prior component of the probability for assigning an observation to an existing cluster in Gibbs sampling. The exact computation depends on the specific process (DP, NGGP, etc.).

Implemented in DP, DPW, NGGP, and NGGPW.

◆ gibbs_prior_existing_clusters()

virtual Eigen::VectorXd Process::gibbs_prior_existing_clusters ( int obs_idx ) const

pure virtual

Compute prior probabilities for assigning observation to all existing clusters.

Parameters

obs_idx Index of the observation to be assigned

Returns: Vector of log prior probabilities for all existing clusters

This method computes the prior components of the probabilities for assigning an observation to each existing cluster in Gibbs sampling. The exact computation depends on the specific process (DP, NGGP, etc.). The returned vector has a length equal to the current number of clusters.

Implemented in DP, DPW, NGGP, and NGGPW.

◆ gibbs_prior_new_cluster()

virtual double Process::gibbs_prior_new_cluster ( ) const

pure virtual

Compute prior probability for creating a new cluster.

Returns: Log prior probability of creating a new cluster

This method computes the prior probability of assigning an observation to a new cluster in Gibbs sampling. The computation is process-specific.

Implemented in DP, DPW, NGGP, and NGGPW.

◆ prior_ratio_merge()

virtual double Process::prior_ratio_merge	(	int	size_old_ci,
		int	size_old_cj ) const

pure virtual

Compute prior ratio for merge move in split-merge algorithm.

Parameters

size_old_ci	Size of first cluster before merge
size_old_cj	Size of second cluster before merge

Returns: Log prior ratio for the merge move

Computes the ratio of prior probabilities when merging two clusters into one cluster. Used in the acceptance probability of merge moves.

Implemented in DP, DPW, NGGP, and NGGPW.

◆ prior_ratio_shuffle()

virtual double Process::prior_ratio_shuffle	(	int	size_old_ci,
		int	size_old_cj,
		int	ci,
		int	cj ) const

pure virtual

Compute prior ratio for shuffle move in split-merge algorithm.

Parameters

size_old_ci	Size of first cluster before shuffle
size_old_cj	Size of second cluster before shuffle
ci	Index of first cluster after shuffle
cj	Index of second cluster after shuffle

Returns: Log prior ratio for the shuffle move

Computes the ratio of prior probabilities when redistributing observations between two existing clusters. Used in restricted Gibbs sampling.

Implemented in DP, DPW, NGGP, and NGGPW.

◆ prior_ratio_split()

virtual double Process::prior_ratio_split	(	int	ci,
		int	cj ) const

pure virtual

Compute prior ratio for split move in split-merge algorithm.

Parameters

ci	Index of first resulting cluster after split
cj	Index of second resulting cluster after split

Returns: Log prior ratio for the split move

Computes the ratio of prior probabilities when splitting a cluster into two clusters. Used in the acceptance probability of split moves.

Implemented in DP, DPW, NGGP, and NGGPW.

◆ set_idx_i()

void Process::set_idx_i ( int i )

inline

Set index of first observation in split-merge pair.

Parameters

i	Index of first observation

◆ set_idx_j()

void Process::set_idx_j ( int j )

inline

Set index of second observation in split-merge pair.

Parameters

j	Index of second observation

◆ set_old_allocations()

void Process::set_old_allocations ( const Eigen::VectorXi & new_allocations )

inline

Store current allocations for potential rollback.

Parameters

new_allocations Current allocation vector to store

Saves the current state of cluster allocations before attempting a split-merge move, enabling rollback if the move is rejected.

◆ update_params()

virtual void Process::update_params ( )

pure virtual

Update process parameters during MCMC sampling.

This method allows derived classes to update their specific parameters during the MCMC chain. For example, some processes may need to sample hyperparameters or update auxiliary variables. Implementation is process-specific.

Implemented in DP, DPW, NGGP, and NGGPW.

Member Data Documentation

◆ data

Data& Process::data

protected

Reference to the data object containing observations and allocations.

◆ idx_i

int Process::idx_i

protected

Index of first observation involved in split-merge move.

◆ idx_j

int Process::idx_j

protected

Index of second observation involved in split-merge move.

◆ log_a

const double Process::log_a = log(params.a)

protected

Precomputed logarithm of total mass parameter for efficiency.

◆ old_allocations

Eigen::VectorXi Process::old_allocations

protected

Storage for previous allocations to enable rollback in case of rejection.

◆ params

Params& Process::params

protected

Reference to the parameters object containing process hyperparameters.

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

src/utils/Process.hpp

Public Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ Process()

◆ ~Process()

Member Function Documentation

◆ gibbs_prior_existing_cluster()

◆ gibbs_prior_existing_clusters()

◆ gibbs_prior_new_cluster()

◆ prior_ratio_merge()

◆ prior_ratio_shuffle()

◆ prior_ratio_split()

◆ set_idx_i()

◆ set_idx_j()

◆ set_old_allocations()

◆ update_params()

Member Data Documentation

◆ data

◆ idx_i

◆ idx_j

◆ log_a

◆ old_allocations

◆ params