Next: Functions in the H2M/cnt Up: Reference Previous: Reference

Subsections


Functions in the main directory


Alphabetical list of functions

c_dgaus
Computes a set of multivariate normal density values in the case of diagonal covariance matrices (mex-file).
gauselps
Plots 2D projections of Gaussian ellipsoids.
gauseval
Computes a set of multivariate normal density values.
gauslogv
Computes a set of multivariate normal log-density values.
hmm
Performs multiple iterations of the EM algorithm.
hmm_chk
Checks the parameters of an HMM and returns its dimensions.
hmm_dens
Reestimates the Gaussian parameters for an HMM.
hmm_fb
Implements the forward-backward recursion (with scaling).
hmm_gen
Generates a sequence of observation given an HMM.
hmm_mest
Reestimates the transition parameters for multiple observation sequences.
hmm_mint
Initializes the distribution parameters using multiple observations (left-right model).
hmm_psim
Generates a random sequence of conditional HMM states.
hmm_tran
Reestimates the transition part of an HMM.
hmm_vit
Computes the most likely sequence of states (Viterbi DP algorithm).
lrhmm
Performs multiple iterations of the EM algorithm for a left-right model.
mix
Performs multiple iterations of the EM algorithm for a mixture model.
mix_chk
Checks the parameters of a mixture model and return its dimensions.
mix_gen
Generates a sequence of observation for a Gaussian mixture model.
mix_par
Reestimates mixture parameters.
mix_post
Computes a posteriori probabilities for a Gaussian mixture model.
randindx
Generates random indexes with a specified probability distribution.
statdis
Returns the stationary distribution of a Markov chain.
svq
Vector quantization using successive binary splitting steps.
vq
Vector quantization using the K-means (or LBG) algorithm.

Notes

The main functions are described in section 2.2 (or in the example scripts), other functions include:

hmm_gen and mix_gen generate data vectors according to a given model. This is useful for testing algorithms on ``prototype data''. hmm_psim generates a random sequence of HMM state conditional to an observation sequence. This can be used for doing Monte Carlo simulations (the way it works is described, for instance, in [10] as ``sampling the indicator variables'').

gauseval and gauslogv compute values of the Gaussian probability density (or the logarithm of it for gauslogv) for several Gaussian distributions and several observed vectors at the same time. Computing as many values as possible at the same time is much faster than calling the function several times (especially when the number of Gaussian distributions is large).

gauselps plots the 2-D projections of the Gaussian ellipsoids corresponding to the Gaussian distribution (this is certainly one of the most useful things in order to see what's going on, at least for low dimensional models).

statdis computes the stationary distribution of a finite state-space Markov chain from its transition matrix.


Next: Functions in the H2M/cnt Up: Reference Previous: Reference

Olivier Cappé, Aug 24 2001