latent: Estimate parameters of a latent variable model

In wrbrooks/latent: Estimate the parameters of a Poisson latent variable model with left-censoring

Description

Estimate the parameters of a latent variable model. Estimation is by the method of maximum likelihood, with conjugate gradient descent used to maximize likelihood. An EM algorithm is used to impute the values for censored observations.

Usage

latent(data, min.detect, event, specific = NULL, verbose = TRUE)

Arguments

data	matrix of observed counts, number of rows equal to the number of observations and columns equal to the number of categories.
min.detect	minimum limit of detection for the FIB assay - values below this threshold are censored
event	vector of event assignment labels, one entry per row of data. Indicates the event to which each row of data belongs.
specific	vector of TRUE/FALSE values, one entry per column of data. Each entry indicates whether the corresponding FIB species is human-specific.
verbose	should the function provide verbose output about its progress? Defaults to TRUE.

Value

A list containing the following elements:

data	final version of the data, where censored values have been imputed by the EM algorithm (see E.step)
min.detect	a vector where each element is the minimum level of detection for the corresponding column of data (copied from function input)
event	vector listing the event ID associated with the corresponding row of the data (copied from the input)
specific	vector indicating whether the corresponding column of the data is a human-specific indicator bacteria - in which case its intercept is set to zero (copied from input)
alpha	matrix of estimated intercept parameters for each indicator bacteria for each event - each row is for one event, and each column is for one species of bacteria
beta	vector of estimated slope parameters for the indicator bacteria, representing the expected marginal increase in log- concentration when the contamination index increases by one unit
gamma	vector of estimated contamination indices, one for each row of the data

Examples

# This script is an example of using the latent package to
# estimate the parameters of a latent variable model for the
# FIB counts in the storm sewer data set (with event 3 removed).

# Import data, drop event 3, change mei[4] from "TNTC" to 0, and then
# convert all FIB to numerics:
data("dfOptAnalysisDataSSJan2015")
indx = which(dfOptAnalysisDataSSJan2015$Event != "03")
fib = dfOptAnalysisDataSSJan2015[indx, c("mei", "modmtec", "FC",
    "Bac.human", "Lachno.2")]
fib$mei[4] = 0
for (n in names(fib))
    fib[[n]] = as.numeric(fib[[n]])

# Set the censoring values for each of the FIB (these are guesstimates)
min.detect = c('mei'=1, 'modmtec'=1, 'FC'=1, 'Bac.human'=225,
    'Lachno.2'=225)

# The human-specific FIB are Bac.human and Lachno.2, which are the fourth
# and fifth columns
specific = c(FALSE, FALSE, FALSE, TRUE, TRUE)

# Get the event IDs
event = as.integer(dfOptAnalysisDataSSJan2015$Event[indx])

# Now estimate the model parameters:
latent(fib, min.detect, event, specific)

wrbrooks/latent documentation built on May 4, 2019, 11:59 a.m.