sample_ghmm.c

EM算法的MATLAB实现,EM算法是HMM中用于随机过程模型参数训练的经典算法

/* sample_ghmm.c 


  Returns samples of a  Multivariate Gaussian process driven by HMM.

 
 Usage  
 -------

  [Z , X]  = sample_ghmm(K , PI , A , M , S , [v1] , ... , [vp] );

 Inputs
 -------

  PI                         Initial state probabilities (d x 1 x [n1] , ... , [nl]) , sum(PI) = ones(1 , 1 , [n1] , ... , [nl] )
  
  A                          State transition probabilities (d x d x [n1] , ... , [nl]), sum(A) = ones(1 , d , [n1] , ... , [nl])
  
  M                          Mean vectors (m x 1 x d x [n1] , ... , [nl])

  S                          Covariance matrices (m x m x d x [n1] , ... , [nl])

 Ouputs
 -------


  Z                          Measurements (m x K x [n1] , ... , [nl] , [v1] , ... , [vp])
  
  X                          State/label sequence (1 x K x [n1] , ... , [nl] , [v1] , ... , [vp])
 
		
		  
  Example
  -------

   d       = 3;
   m       = 2;
   N       = 1;
   K       = 3;
   L       = 4;
 
   PI      = rand(d , 1 , N);
   sumPI   = sum(PI);
   PI      = PI./sumPI(ones(d , 1) , : , :);


   A       = rand(d , d);
   sumA    = sum(A);
   A       = A./sumA(ones(d , 1) , : , :);


   M       = 3*randn(m , 1 , d);
   S       = cat(3 , [2 , 1.1 ; 1.1 1.5] , [1.3 , 0.3 ; 0.3 1.5] ,  [1 , 0 ; 0 1]);


   %[Z , X] = sample_ghmm(K , PI(: , : , ones(1 , 4) ,  ones(1 , 2)) , A(: , : , ones(1 , 4), ones(1 , 2)) , M(: , : , : , ones(1 , 4), ones(1 , 2)) , S(: , : , : , ones(1 , 4), ones(1 , 2)) );


   [Z , X] = sample_ghmm(K , PI , A , M , S , L);


   %[Z , X] = sample_ghmm(K , PI(: , : , ones(1,L)) , A(: , : , ones(1,L)) , M(: , : , : , ones(1,L)) ,  S(: , : , : , ones(1,L)));

  
 
				
  To compile
  ----------
  
  mex  -DranSHR3 -output sample_ghmm.dll sample_ghmm.c

  
  Myself, I use Intel CPP compiler as : 

  
  mex -DranSHR3 -f mexopts_intel10amd.bat -output sample_ghmm.dll sample_ghmm.c

  Ver 1.0 (03/16/05)


  Author : S閎astien PARIS