Main Page   Compound List   File List   Compound Members   File Members   Related Pages  

ParaSearch Class Reference

object containing a collection of Multinomial distributions updated by dirichlet model hyper parameters used by GraphManipulator to learn the parameter vector of the similarity model. More...

#include <ParaSearch.hh>

List of all members.

Public Methods

 ParaSearch ()
virtual ~ParaSearch ()
void Allocate (int nb_parameters, int nbLevel, int nb_MTClasses, int nb_SubGMTClasses, int min_Sample, int max_Sample, int nb_TimeSamples)
 memory allocation method.

void Init (SearchTree ****searchTreeCollec, float ****costss, vector< int > *kMTa, vector< int > *kMTb, vector< int > *firstSample)
 Initialisation of the different pointers : to the tree structures, the costs collection and the a posteriori collection of sub-graphs.

void Init (vector< int > *kMTa, vector< int > *kMTb, vector< int > *firstSample)
 Initialisation of the pointers the a posteriori collection of sub-graphs.

void MaxCostalues ()
 Calculation method of the maximum values of the matching costs.

void MMSEestim (int interestkMTa, int interestkMTb, int interestfirstSample)
 Method to perform a MMSE estimate of the parameter vector of the similarity model.

float ReevaluateMatch (int i, int j, int k)
 Method to update of the similarity with a new parameter vector. More...

void ErasePdf ()
 Method use to transfom into a flat distribution the learned multinomial distributions.


Public Attributes

int nbPara
 number of elements in the parameter vector of the similarity model.

int nbQuantifLevel
 number of bins used for the interactive estimation of the multinomial distributions.

int nbMTClasses
 number of MT classes composing the global graph.

int nbSubGMTClasses
 number of MT classes considered per sub-graph in the matching procedure.

int minSample
 Minimum time sample considered in the global graph.

int maxSample
 Maximum time sample considered in the global graph.

int nbTimeSamples
 number of samples of the time-window.

SearchTree **** treeCollec
 pointer to collections of SearchTree objects.

float **** costs
 pointer to collections of distorsion costs array. More...

vector< int > kMTaSorted
 vector of MT class indexes attached to the elements of the a posteriori collection of sub-graphs.

vector< int > kMTbSorted
 vector of 2nd MT class indexes attached to the elements of the a posteriori collection of sub-graphs (only when 2 MT classes are considered simultaneously per sub-graph).

vector< int > firstSampleSorted
 array of first time sample indexes attached to the elements of the a posteriori collection of sub-graphs.

float ** pdf
 array of multinomial distributions for approximating the paramter vector of the similarity model PDF.

float * maxPdf
 array of maximum values of the multinomial distributions.

float * meanPdf
 array of mean values of the multinomial distributions.

float * estimPara
 array of MMSE parameter estimates of the similarity model.

float * estimParaPrevious
 array of MMSE parameter estimates of the similarity model at the previous learning iteration.

float * maxCostValue
 array of maximum cost values.


Detailed Description

object containing a collection of Multinomial distributions updated by dirichlet model hyper parameters used by GraphManipulator to learn the parameter vector of the similarity model.

For this, the Multinomial distributions, which are the parameter distributions learned by the system, are used to evaluate with MMSE estimator the currently learned parameter vector of the similarity model.

Definition at line 32 of file ParaSearch.hh.


Member Function Documentation

float ParaSearch::ReevaluateMatch int    i,
int    j,
int    k
[inline]
 

Method to update of the similarity with a new parameter vector.

This methods calls the SearchTree object method 'LowestCosts(...)'which updats the matching costs with a new parameter vector while using the already created tree structure. It returns the updated costs related to the contribution of the different graph's attribute for the updated best mapping function (path of minimal cost)

Definition at line 187 of file ParaSearch.hh.


Member Data Documentation

float**** ParaSearch::costs
 

pointer to collections of distorsion costs array.

The cost assigned during the matchingprocedure to the best mapping function is decomposed into several parts related to each attribute contribution. These different cost constribution are stored in the array 'costs'.

Definition at line 57 of file ParaSearch.hh.


The documentation for this class was generated from the following file:
Generated on Thu Feb 17 11:03:20 2005 for Interactive Learning of Sub-Graphs Semantics by doxygen1.2.14 written by Dimitri van Heesch, © 1997-2002