Parameters for HMM-VB training.
This function creates a list with parameters for estimating an HMM-VB,
which is used as an argument for hmmvbTrain.
trainControl(ninit0 = 1, ninit1 = 0, ninit2 = 0, epsilon = 1e-04, diagCov = FALSE)
ninit0 |
The number of initializations for default scheme 0, under which the k-means clustering for entire dataset is used to initialize the model. |
ninit1 |
The number of initializations for default scheme 1, under which the k-means clustering for a subset of data is used to initialize the model. |
ninit2 |
The number of initializations for default scheme 2, under which a random subset of data is used as cluster centroids to initialize the model. |
epsilon |
Stopping criteria for Baum-Welch algorithm. Should be a small number in range (0,1). |
diagCov |
A logical value indicating whether or not variable block covariance matrices will be diagonal. |
The named list with parameters.
# setting up multiple initialization schemes
Vb <- vb(1, dim=4, numst=2)
set.seed(12345)
hmmvb <- hmmvbTrain(iris[,1:4], VbStructure=Vb,
trControl=trainControl(ninit0=2, ninit1=2, ninit2=2))
show(hmmvb)
# forcing diagonal covariance matrices
Vb <- vb(1, dim=4, numst=2)
set.seed(12345)
hmmvb <- hmmvbTrain(iris[,1:4], VbStructure=Vb,
trControl=trainControl(diagCov=TRUE))
show(hmmvb)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.