Emmerhond Emmerhond - 1 month ago 19
R Question

Markov switching model, R package, how to view source code

I am trying to look into the source code of an implementation of a general Markov switching model, in the MSwM package. I'm new to R, and I already have managed to look into the function msmFit().

However there is another function inside msmFit(), at the bottom called em(). I have not yet been able to view the source code of this inside function. Does anyone know how to view this em()-function?

Many thanks

Code and output



> getMethod(msmFit, signature = c(object="lm", k="numeric", sw="logical", p="ANY", data="missing", family="missing")):


Method Definition:

function (object, k, sw, p, data, family, control)
{
if (!missing(data)) {
if (is.list(data)) {
if (class(data[[1]]) == "call") {
call = data[[1]]
}
else {
call = match.call()
}
}
else {
call = match.call()
}
}
else {
call = match.call()
}
if (missing(p))
p = 0
if (missing(control))
control = list()
control <- do.call(msmControl, control)
if (p > 0) {
var = object$model[, 1]
Ar = apply(as.matrix(1:p), 1, function(el) {
length(var) = length(var) - el
var = c(rep(NA, el), var)
return(var)
})
colnames(Ar) = paste(names(object$model)[1], "_", 1:p,
sep = "")
aux = paste(colnames(Ar), collapse = "+")
object = update(formula = as.formula(paste("~.+", aux,
sep = "")), data = data.frame(object$model, Ar),
object)
}
Coef = data.frame(matrix(NA, nrow = k, ncol = length(coef(object))))
std = rep(0, k)
ind = sample(1:k, length(object$residuals), replace = T)
for (i in 1:k) {
data1 = as.data.frame(object$model[ind == i, , drop = F])
mod1 = update(object, formula = object$terms, data = data1)
Coef[i, ] = coef(mod1)
std[i] = summary(mod1)$sigma
}
names(Coef) = names(coef(object))
transMat = t(matrix(table(ind, c(ind[-1], NA))/rep(table(ind[-length(ind)]),
k), ncol = k))
ans = new(Class = "MSM.lm", call = as.call(call), model = object,
k = k, switch = sw, p = p, Coef = Coef, std = std, transMat = transMat,
iniProb = rep(1/k, k))
validMSM.linear(ans)
validMSM.lm(ans)
ans = em(ans, control)
return(ans)
}
<environment: namespace:MSwM>

Signatures:
object k sw p data family
target "lm" "numeric" "logical" "ANY" "missing" "missing"
defined "lm" "numeric" "logical" "ANY" "missing" "missing"

Answer

If you're using the conventional R stack, git clone the MSwM project to your own machine and look in the R subdirectory. It looks like em is a generic (I'm not an expert on the R object models!) which is why it has a funny extended name. You may be able to get away with looking directly at the body() of .MSM.em, but I'm not sure if it will be visible from your R session.

Please consider changing the tags for this question as the technical aims of the package (as opposed to the details of code structure!) aren't very relevant.