Model Parameter Estimation

Estimates the GHMM's parameters by maximizing the likelihood of a given observation sample, using a variation of the Baum-Welch algorithm[#!em_tut!#]. The reestimation equations used can be found in [#!fernando2006!#].

Usage: training( $<$it$>$ [,$<$thr$>$], $<$object$>$ )

where $<$it$>$ is the maximum number of iterations that the Baum-Welch algorithm will perform before stopping; $<$thr$>$, an optional parameter, is the log-likelihood gain^11.7threshold that, when reached, will stop the estimation algorithm; and $<$object$>$ is the MTK object containing the observation sample.

Output: Shows the progress of the estimation, by printing the current iteration and the likelihood of the observation sample used in the training, given the current parameter estimates. The values estimated for the parameters are automatically stored in the ghmm object's attributes.

Optimized Parameter Estimation: Besides the training() method, the GHMM plugin also implements another parameter estimation method, called training_fast(). The difference between both is that while the first one calculates, for each observation in the sample, a new set of parameters for the model, the second one calculates, in a single step, these new parameters for a hole observation batch, thus improving its estimation speed by a factor of $B$[#!fernando2006!#]. The parameters taken by this method, and its output, are identical to those of the training() method:

Usage: training_fast( $<$it$>$ [,$<$thr$>$], $<$object$>$ )