loglikelihood: Calculates the log-likelihood value

In nvihrs14/tcherry: Learning the structure of tcherry trees

Description

Calculates the value of the log-likelihood function for a graphical model with the given junction tree after inserting estimated probabilities.

Usage

loglikelihood(cliques, separators, data, ...)

Arguments

cliques	A list containing the cliques of the junction tree.
separators	A list containing the separators of the junction tree.
data	A data frame with observations of the variables.
...	Additional arguments passed to weight_junction_tree.

Details

Calculates the value as

n_obs * (weight - ∑ H(A))

where n_obs is the number of observations in data, weight is the estimated weight of the junction tree and H is the estimated entropy of the variable A. The sum is over all variables in data. All estimated probabilities are maximum likelihood estimates unless the smooth argument has been used to deal with zero probabilities.

Value

The calculated log-likelihood value.

Author(s)

Katrine Kirkeby, [e-mail]

Maria Knudsen, [e-mail]

Ninna Vihrs, [e-mail]

See Also

weight_junction_tree for calculating the weight of a junction tree.

Examples

set.seed(43)
var1 <- c(sample(c(1, 2), 100, replace = TRUE))
var2 <- var1 + c(sample(c(1, 2), 100, replace = TRUE))
var3 <- var1 + c(sample(c(0, 1), 100, replace = TRUE,
                        prob = c(0.9, 0.1)))
var4 <- c(sample(c(1, 2), 100, replace = TRUE))
var5 <- var2 + var3
var6 <- var1 - var4 + c(sample(c(1, 2), 100, replace = TRUE))
var7 <- c(sample(c(1, 2), 100, replace = TRUE))

data <- data.frame("var1" = as.character(var1),
                   "var2" = as.character(var2),
                   "var3" = as.character(var3),
                   "var4" = as.character(var4),
                   "var5" = as.character(var5),
                   "var6" = as.character(var6),
                   "var7" = as.character(var7))

cliques <- list(c("var1", "var2", "var3"),
                c("var2", "var3", "var5"),
                c("var5", "var6", "var7"),
                c("var1", "var4"),
                c("var2", "var5", "var6"))

separators <- list(c("var1"),
                   c("var2", "var3"),
                   c("var2", "var5"),
                   c("var5", "var6"))

# smooth is used to deal with zero probabilities.
loglikelihood(cliques, separators, data, smooth = 0.1)

nvihrs14/tcherry documentation built on Aug. 1, 2020, 6:25 p.m.