JaCHMM Files

This archive contains the JaCHMM library - a Java-based implementation of a Conditioned Hidden Markov Model. It is based on the Jahmm library.
Right now, only the implementation containing label independent tranistions is available.

Here is a brief example of how to use the library from the command line with multi_gaussian output density function function:

Creating dummy JaCHMM file:
java ulm.ce.ds.jachmm.apps.cli.Cli create -opdf multi_gaussian -n $nbLabels -nh $nbHiddenStates -d $DIMENSION -o $INITIAL_CHMM_FILE

Initilaizing the CHMM using K-Means:
java ulm.ce.ds.jachmm.apps.cli.Cli learn-kmeans -opdf multi_gaussian -d $DIMENSION -is $TRAINING_VECTOR_FILE_NAME -i $INITIAL_CHMM_FILE -o $KMEANS_FILE_NAME -oc $OPTIMIZE_CLUSTERS -lf $TRAINING_LABEL_FILE_NAME -km $KMEANS_MODE -ni $nbIterationsKM

Training using Baum-Welch:
java ulm.ce.ds.jachmm.apps.cli.Cli learn-bw -opdf multi_gaussian -d $DIMENSION -i $KMEANS_FILE_NAME -is $TRAINING_VECTOR_FILE_NAME -o $BW_FILE_NAME -ni $nbIterationsBW

Performing Viterbi:
java ulm.ce.ds.jachmm.apps.cli.Cli va -opdf multi_gaussian -d $DIMENSION -is $EVALUATION_VECTOR_FILE_NAME -i $KMEANS_FILE_NAME
(file mostLikelyLabels.seq is generated)

Estimating most probable label:
java ulm.ce.ds.jachmm.apps.cli.Cli label-est -opdf multi_gaussian -d $DIMENSION -i CHMM_FILE -is $EVALUATION_VECTOR_FILE_NAME