fast_E_M: EM clustering

In svs: Tools for Semantic Vector Spaces

Description

A fast procedure for Expectation-Maximization clustering.

Usage

fast_E_M(dat, k, tol = 1e-08)

fast_EM(dat, k, tol = 1e-08)

Arguments

dat	Input data: can be a table or a data frame (but the data frame must have only two columns).
k	Numeric specification of the number of latent classes to compute.
tol	Numeric specification of the convergence criterion.

Details

This function assumes that the rows of a frequency table come from a multinomial distribution. The prior probabilities of the latent classes are initialized with a Dirichlet distribution (by means of rdirichlet from the package gtools) with alpha = the total frequency counts of every level.

Value

A list with components:

prob0	The probabilities of the latent classes.
prob1	The probabilities for the first set of levels (viz. the row levels of a frequency table). The rows of prob1 sum to 1.
prob2	The probabilities for the second set of levels (viz. the column levels of a frequency table). The rows of prob2 sum to 1.

References

Dempster, A. P., N. M. Laird and D. B. Rubin (1977) Maximum likelihood from incomplete data via the EM algorithm. Journal of the royal statistical society, series B 39 (1), 1–38.

Examples

SndT_Fra <- read.table(system.file("extdata", "SndT_Fra.txt", package = "svs"),
   header = TRUE, sep = "\t", quote = "\"", encoding = "UTF-8",
   stringsAsFactors = FALSE)
E_M.SndT_Fra <- fast_E_M(SndT_Fra, k = 7)
E_M.SndT_Fra

svs documentation built on Nov. 10, 2020, 1:09 a.m.