src/hmm.h File Reference

STATUS: work in progress Hidden Markov Models. More...

#include "mathadd.h"
#include "definitions.h"
#include "eeg.h"
#include <gsl/gsl_rng.h>
#include <gsl/gsl_randist.h>

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Data Structures
struct	CPHiddenMarkovModel
Functions
CPHiddenMarkovModel *	cphmm_alloc (int K, int n, int M, int Q, int J)
void	cphmm_free (CPHiddenMarkovModel *m)
double	cphmm_get_transition_prob (CPHiddenMarkovModel *m, int k, int i, int j)
void	cphmm_init (CPHiddenMarkovModel *m, double **X)
CPHiddenMarkovModel *	eeg_cphmm_init (EEG *eeg, int channel, double **X)

---

Detailed Description

STATUS: work in progress Hidden Markov Models.

Warning:

For now this is only experimental but plans are to complete this file.

Implements the continuous profile model from:

Listgarten et al. Multiple alignment of continuous time series. Advances in Neural Information Processing Systems (2005) vol. 17 pp. 817–824

Definition in file hmm.h.

---

Function Documentation

CPHiddenMarkovModel* cphmm_alloc	(	int	K,
		int	n,
		int	M,
		int	Q,
		int	J
	)

allocate memory for the CPM.

Parameters:

	K	number of trials
	n	number of samples per trial
	M	number of samples in latent trace
	Q	number of amplitude states
	J	maximum number of jumps in warping

Definition at line 55 of file hmm.c.

void cphmm_free

(

CPHiddenMarkovModel *

m

)

deallocate the model's struct. RNG is dealloc'ed as well.

Definition at line 165 of file hmm.c.

double cphmm_get_transition_prob	(	CPHiddenMarkovModel *	m,
		int	k,
		int	i,
		int	j
	)

STATUS: work in progress return state transition probabilities from state i to state j in trace k:

where

Details see Listgarten et al. 2005

Todo:

not implemented yes

Definition at line 36 of file hmm.c.

void cphmm_init	(	CPHiddenMarkovModel *	m,
		double **	X
	)

initializing the model according to the parameters from Listgarten et al. 2005:

• sigma = 15% of max(X[0])-min(X[0])
• z = first observed time-series + gaussian noise (0, sigma) upsampling to match M
• u = (1,...,1)
• d,s = uniform
• q is filled with Q numbers evenly spaced in log-space [0,2]

  Parameters:

  	m	the model
  	X	the data, K x n

Definition at line 97 of file hmm.c.

CPHiddenMarkovModel* eeg_cphmm_init	(	EEG *	eeg,
		int	channel,
		double **	X
	)

init the model from EEG-data. The data matrix X is a return value.

Parameters:

	eeg
	channel	- which channel is used?
	X	- the converted data. X should contain K empty pointers. These are set to the appropriate data in eeg (not copied).

Definition at line 140 of file hmm.c.