me.VEV {mclust} | R Documentation |
EM iteration (M-step followed by E-step) for estimating parameters in an MVN mixture model having constant shape, varying volume and possibly one Poisson noise term.
me.VEV(data, z, eps, tol, itmax, equal = F, noise = F, Vinv)
data |
matrix of observations. |
z |
matrix of conditional probabilities. z should have a row for each observation
in data , and a column for each component of the mixture.
|
eps |
A 2-vector specifying lower bounds on the pth root of the volume of the
ellipsoids defining the clusters, where p is the data dimension, and on the
reciprocal condition number for the estimated shape of the covariance
estimates. Default : c(.Machine$double.eps, .Machine$double.eps) .
If only one value is given it is assumed to override only the first
default.
|
tol |
A 2 vector giving the tolerances for the outer (EM) and inner (volume and
shape estimating) iterations. The outer iteration is terminated if the
relative error in the loglikelihood value falls below tol[1] .
The inner iteration is terminated if the relative error in both the estimate
of the pth root of the volume and the shape estimate falls below tol[2] .
Default: c(sqrt(.Machine$double.eps), sqrt(.Machine$double.eps)) .
If only one value is given it is assumed to override only the first
default.
|
itmax |
A 2-vector giving an upper limit on the number of outer and inner iterations.
Default : c(Inf,Inf) - no upper limit for outer or inner iterations.
If only one value is given it is assumed to override only the first
default.
|
equal |
Logical variable indicating whether or not to assume equal proportions in the
mixture. Default : F .
|
noise |
Logical variable indicating whether or not to include a Poisson noise term in
the model. Default : F .
|
Vinv |
An estimate of the inverse hypervolume of the data region (needed only if
noise = T ). Default : determined by function hypvol
|
the conditional probablilities at the final iteration (information about the iteration is included as attributes).
The default for inner iterations are set up so as to compute the true M-step
parameters at each iteration. However if you plan to run me.VEV
to
convergence, then it is usually safe to set the number of inner iterations to
0, 1, or some small number.
G. Celeux and G. Govaert, Gaussian parsimonious clustering models, Pattern Recognition, 28:781-793 (1995).
A. P. Dempster, N. M. Laird and D. B. Rubin, Maximum Likelihood from Incomplete Data via the EM Algorithm, Journal of the Royal Statistical Society, Series B, 39:1-22 (1977).
G. J. MacLachlan and K. E. Basford, The EM Algorithm and Extensions, Wiley, (1997).
data(iris) cl <- mhclass(mhtree(iris[,1:4]),3) z <- me.VEV( iris[,1:4], ctoz(cl)) mstep.VEV(iris[,1:4], z)