Você está na página 1de 8

Continuous Emission Hidden Markov Models for Sequence Classification

Pi19404
March 13, 2014

Contents

Contents
Hidden Markov Models with continuous emission
0.1 Introduction . . . . . . . . . . . . . . . 0.2 Continuous Emission . . . . . . . . 0.2.1 Gaussian Mixture Models 0.2.2 Forward Algorithm . . . . 0.3 Code . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 3 5 5 8 8

2|8

Hidden Markov Models with continuous emission

Hidden Markov Models with continuous emission


0.1 Introduction
In this article we will look at hidden Markov models with continuous emission and its application in classification of D dimensional sequential data . The emission probabilities are modeled using Gaussian mixture model.The implementation is developed as part of OpenVision -Vision and Machine learning Library.
 Let us assume that observed process Y ; n 2 N is vector valued in a euclidean space O  R and X ; n 2 N are the hidden states.
n d n

 The pair of processes Markov model.

X and Y is called a continuous hidden


n n

0.2 Continuous Emission


 In the case of discrete hidden Markov model we had considered the emission to be discrete random variance and specified a emission matrix.  In the case of continuous hidden Markov model the observed random variables are continuous in nature.Let the continuous random variable be defined in a dimensional Euclidean space. Y = y1 ; : : : ; y is the observed random variable.

 At each time step the system generates a hidden state cording to a a state to state transition matrix T .
i;j t

X act

 Once x has been generated the system generates a hidden cluster y according to the state to emission probability distribution.
t

 Gaussian mixture models are used to model a multi-modal density function in Euclidean space.  Each hidden state is associated with emission probability distribution.

3|8

Hidden Markov Models with continuous emission


 The emission probability distribution are specified in terms of conditional distribution conditioned on the hidden state.  The HMM can be viewed as a parametric model whose parameters vary with the state of the underlying hidden Markov chain.  Let N be number of hidden states ,X

= x1; : : : ; x ; n 2 N
n

 Let M be the number of mixture components

P (yjx ) =
Y i

X=1
k

a ( ;
i;k i;k

i;k

i;k

is a vector of dimension

i;k

is a matrix of size

DxD

 Given we have a observation Y we need to determine the probability that observation is emitted by the hidden state X that is given by the modeled conditional distribution.  This distribution is used to model the emission probability.  Thus the continuous emission hidden Markov model is specified by transition matrix,initial probability matrix and emission probability distribution specified by Gaussian mixture model.

MatrixXf _transition; MatrixXf _initial; vector<GaussianMixture> _emission; int _nstates; int _nobs;
 Let M be the number of hidden states,possible discrete value the random variable X can take.
n

 Let

 = (; A; B ) represent hidden Markov model


p y M

 With a continuous HMM B corresponds to a family of parametric distributions p () where  2  R and B = (1 ; : : : ;  ).  For each hidden state we have a associated parametric distribution.

P (y;  ) =
Y i

Xa
k

i;k

 ( ;
i;k

i;k

 Another interpretation of above representation is,that hidden state i consists of hidden subsets k 2 N

4|8

Hidden Markov Models with continuous emission


 And each of these subsets k can generate the observed symbol Y with probabilities a and probabilities that observed symbol is generate from specific hidden state is given by  ( ; ).Thus total probability can be specified as a product a  ( ; ),and average probability that observed symbol is generate by the hidden state i is given by the mixture model.
i;k i;k i;k i;k i;k i;k

 Again the task is the present article is sequence classification.  For the implementation we require implementation of Gaussian Mixture model and then see the modifications required in the forward/backward algorithm for continuous emission models.

0.2.1 Gaussian Mixture Models


 The mixture models are specified by number of mixture components and parameters for each components k 2 M and parameter of mixture components  ( ; )
i;k i;k

 We can consider this to be a soft-clustering process where a continuous random vector x is assigned to clusters specified with the mixture models with some probability.
n

 The clusters can be considers as hidden discrete emission which again emits a continuous random vector which is truly being observed.  Thus we can consider than we have a another hidden latent variable which is the discrete emission specified by the cluter.  The Class GaussianMixture which can be found in OpenVision Library which encapsulates Gaussian Mixture model is used to model the emission probabilities

0.2.2 Forward Algorithm


 The goal in the forward algorithm is to compute the probability

P (y1 ; : : : ; y

 this was computed recursively using

(z

) = P (x1; : : : ; x ; z ) (z ) =
n n n n n

X (Z
n1

1 )P (x jz )P (z jz 1 )
n n n n

 The quantity P (x jz ) was take from emission matrix in case of discrete hidden Markov model.

5|8

Hidden Markov Models with continuous emission


 In case of continuous HMM this is required to be computed from the parametric distribution P (x = xjz = z ) = a  ( ;
n n i k i;k i;k

i;k

 Remaining computation remain the same,just instead of a lookup table in case of discrete distribution we have to perform computation according to model of parametric distribution in case of continuous observation variable.  To verify the code the scikit-learn -a ptyhon machine learning package was used consider the following hidden Markov model for sequence classification.

#number of hidden states _nstates=2; #number of emissions per state _nmix=2; #length of the feature vector ndim=3; #transition probability model.transmat_=[[0.9,0.1],[0.1,0.9]]; #inital state probability model.startprob=[0.5,0.5]; #multivariate Gaussian for emission 1 a1=GMM(n_components=_nmix,covariance_type='full'); #the mean vector a1.means_=np.array([[ 0.08513302,-0.02109036,-0.54525889], [-0.02646832,0.01030935,0.55921764]]); #the covariance vector a1.covars_=np.array([[ [1.6045371 ,0,0],[0, 0.95491166,0],[0,0, 0.76062784]] [ [0.63467475,0,0],[0, 1.03482635,0],[0,0, 0.53010913]]]); #the mixture weights a1.weights_=[0.48161747,0.51838253];

#multivariate Gaussian for emission 2 a2=GMM(n_components=_nmix,covariance_type='full'); a2.means_=np.array([[ 0.06263125, -0.01238598, -0.38820614], [ 0.3781768 , -0.29539352, 0.84720285]]); a2.covars_=np.array([[ [0.66081899, 0,0],[0,1.3027587,0],[0,0, 0.68465955]], [ [1.23561827,0,0],[ 0, 0.55178899,0],[ 0,0,0.80040462]]]); a2.weights_=[ 0.74733838, 0.25266162]; model.gmms_[0]=a1; model.gmms_[1]=a2; #input feature vector sequence x=[[0.3,0.4,0.5],[0.1,0.2,0.3],[0.4,0.4,0.6]];

6|8

Hidden Markov Models with continuous emission

#predicting the probability that sequence belongs to model print model.score(x)


 The libconfig++ library is used to load the parameters of the HMM from the configuration files.  A sample code for testing the Continuous emission HMM is shown below

int hmm_test1(int argc,char **argv) { Config cfg; // Read the file. If there is an error, report it and exit. try { cfg.readFile(argv[1]); } catch(const FileIOException &fioex) { std::cerr << "I/O error while reading file." << std::endl; return(EXIT_FAILURE); } catch(const ParseException &pex) { std::cerr << "Parse error at " << pex.getFile() << ":" << pex.getLine() << " - " << pex.getError() << std::endl; return(EXIT_FAILURE); } //defining the object for continuous emission hmm ocv::CHMM hmm1; //load the parameters of hmm hmm1.loadConf(cfg,"hmm1"); Mat sequence=Mat(3,3,CV_32FC1); sequence.at<float>(0,0)=0.3; sequence.at<float>(0,1)=0.4; sequence.at<float>(0,2)=0.5; sequence.at<float>(1,0)=0.1; sequence.at<float>(1,1)=0.2; sequence.at<float>(1,2)=0.3; sequence.at<float>(2,0)=0.4; sequence.at<float>(2,1)=0.4; sequence.at<float>(2,2)=0.6;

7|8

Hidden Markov Models with continuous emission

float l=hmm1.likelyhood(sequence); cerr << "kkkkkkkkk" << l <<endl;

 The result estimated from the developed code is in accordance with the results provided by the scikit-learn package.

0.3 Code
The class CHMM encapsulates the continuous emission hidden Markov model.The class can be found in files hmm.cpp and hmm.hpp files in the OpenVision :A Machine vision library at https:// github.com/pi19404/OpenVision in ImgML subdirectory. The configuration files can be found in the conf subdirectory of the repository.

8|8

Você também pode gostar