EM Control {EMCluster}

EM Control Generator and Controller

Description

The .EMControl generates an EM control (.EMC) controlling the options and conditions of EM algorithms, i.e. this function generate a default template. One can either modify .EMC or employ this function to control EM algorithms. By default, .EMC, .EMC.Rnd, and .EC.Rndp are three native controllers as the EMCluster is loaded.

Usage

.EMControl(alpha = 0.99, short.iter = 200, short.eps = 1e-2,
    fixed.iter = 1, n.candidate = 3,
    em.iter = 1000, em.eps = 1e-6, exhaust.iter = 5)
.EMC
.EMC.Rnd
.EMC.Rndp

Arguments

alpha

only used in emgroup for "SVD" initialization.

short.iter

number of short-EM steps, default = 200.

short.eps

tolerance of short-EM steps, default = 1e-2.

fixed.iter

fixed iterations of EM for "RndEM" initialization, default = 1.

n.candidate

reserved for other initialization methods (unimplemented).

em.iter

maximum number of long-EM steps, default = 1000.

em.eps

tolerance of long-EM steps, default = 1e-6.

exhaust.iter

number of iterations for "exhaustEM" initialization, default = 5.

Details

exhaust.iter and fixed.iter are used to control the iterations of initialization procedures.

short.iter and short.eps are used to control the short-EM iterations.

em.iter and em.eps are used to control the long-EM iterations.

Moeover, short.eps and em.eps are for checking convergence of the iterations.

Value

This function returns a list as .EMC by default.

The .EMC.Rnd is equal to .EMControl(short.eps = Inf) and usually used by the rand.EM method.

The .EMC.Rndp is equal to .EMControl(fixed.iter = 5) where each random initials run 5 EM iterations in the rand.EM method.

Author(s)

Wei-Chen Chen [e-mail] and Ranjan Maitra.

References

https://www.stat.iastate.edu/people/ranjan-maitra

See Also

init.EM, emcluster.

Examples

library(EMCluster, quietly = TRUE)

.EMC <- .EMControl()
.EMC.Rnd <- .EMControl(short.eps = Inf)
.EMC.Rndp <- .EMControl(fixed.iter = 5)