treeHFM (version 1.0.3)

HFMfit: Fit a HFM Model

Description

The functions fits an HFM (Hidden Factor Graph Model) to the (tree structured) data.

Usage

HFMfit(data,nHStates,dataNodeIndices,priorInit,transmatInit,transInitDiv,
	type,emissionProb=c(),SigmaInit=c(), muInit=c() )

Arguments

data

The observations. A list of measurements for every observation. An entry can be a matrix of size KxN, whereas K is the dimension of the data and N the lenght of the sequence or of size 1xK whereas the entries are discrete.

nHStates

Number of hidden states H

dataNodeIndices

The data structure. A list of the toplogy of every data entry of the format [parentIndex childIndex].

priorInit

The inital state probabilities

transmatInit

The initial transtition probabilities for sequential transitions (H x H)

transInitDiv

The initial transtition probabilities for splitting events (H x (H*H)

type

Type of the data. ('d' discrete or 'c' continous)

emissionProb

Initial emission probilities for discrete data, a matrix of size E x H, whereas E are the number of discrete observable states.

SigmaInit

Initial covariance matrix for continous data

muInit

Initial means for continous data

Value

An HFM: An HFM:

Details

The algorithm fits an HFM to sequential data (in this case essentially a Hidden Markov Model is fitted) or to data that is structured as a binary tree. An HFM consists of two transition probability matrices. One for the sequential transitions and one for transitions where a splitting happens.

Examples

## Fit a discrete treeHFM ## 
nOStates = 3;
nHStates = 2;
#
T = 1;
nex = 10;
obs1=sample(1:nOStates,5,replace=T)
obs2=sample(1:nOStates,5,replace=T)
data=list()
data[[1]]=obs1;
#
nodeIndices1=cbind(c(0,1,2,2,4),c(1,2,3,4,5));
dataNodeIndices=list()
dataNodeIndices[[1]]=nodeIndices1;
prior1=array(1,nHStates);
priorInit = array(1,nHStates)/nHStates;
transmatInit = matrix(1,nHStates,nHStates)*(1/nHStates);
transInitDiv= matrix(1,nHStates,nHStates*nHStates)*(1/(nHStates*nHStates))
obsmatInit = matrix(sample(1:nOStates,nHStates*nOStates,replace=T),nHStates,nOStates);
obsmatInit=obsmatInit/rowSums(obsmatInit,1);
hfm=HFMfit(data,nHStates,dataNodeIndices,priorInit,transmatInit,transInitDiv,
'd',emissionProb=obsmatInit);
## Fit a continous treeHFM ## 
nHStates = 2;
########create observation sequences########
obs1 <- rbind(runif(10,0,1),runif(10,0,1))
obs2 <- rbind(runif(8,0,1),runif(8,0,1))
data=list()
data[[1]]=obs1
data[[2]]=obs2
######### create guesses for gaussian covariance matrix and means #########
mc2 <- Mclust(t(cbind(obs1,obs2)), G=2)
muInit=mc2$parameters$mean;
SigmaInit=mc2$parameters$variance$sigma
#########create tree topology####################################
nodeIndices1=cbind(c(0,1,2,3,3,4,5,6,7,8),c(1,2,3,4,5,6,7,8,9,10));
nodeIndices2=cbind(c(0,1,2,3,4,4,5,6),c(1,2,3,4,5,6,7,8));
dataNodeIndices=list()
dataNodeIndices[[1]]=nodeIndices1;
dataNodeIndices[[2]]=nodeIndices2;
######### create guesses for prior and transition matrices#########
prior1=array(1,nHStates);
priorInit = array(1,nHStates)/nHStates;
transmatInit = matrix(1,nHStates,nHStates)*(1/nHStates);
transInitDiv= matrix(1,nHStates,nHStates*nHStates)*(1/(nHStates*nHStates))
#
hfm=HFMfit(data,nHStates,dataNodeIndices,priorInit,transmatInit,transInitDiv,'c',
SigmaInit=SigmaInit, muInit=muInit);
#