How to use predict() function in Markov chain

Question

I am trying to create a function sing Markov Chain to make future predictions, but it's giving me a warning message. Please help.

My R version is 3.6.1.

Answer 1

The Problem

The issue you're having is that you're passing a vector of length > 1 to the argument t0 in rmarkovchain() , when it expects t0 to be of length exactly one. Think about your predict function:

predict <- function(initial_state, t) {
    prediction <- rmarkovchain(n = t, object = mcWeather, t0 = initial_state)[t] 
    prediction 
}

When a vector of length greater than one is passed as the argument initial_state , then rmarkovchain() gets a value for t0 that is more than one element, which causes problems with markovchainSequence() , which it calls.

Why does this cause a problem? We can see from help("markovchainSequence") that it expects t0 is of length one:

Arguments

...[Some content omitted]...

t0 The initial state

(emphasis added). So, when it does some checking about the t0 value, it uses if() , which expects a logical vector of only length one. That's why when it checks

if (!(t0 %in% markovchain@states)) 

if you've passed a vector of length more than one as t0 , you get a warning; the if() function expects only one logical value, so it will use only the first logical value in the resulting vector. That's why you get the warning, it's also why all your predictions are the same.

The Solution

So, how do we fix it? We alter your function slightly to handle vectors of length greater than one for initial_state :

predict <- function(initial_state, t) {
    prediction <- sapply(initial_state, function(state) {
        rmarkovchain(n = t, object = mcWeather, t0 = state)[t]
    })
    return(prediction)
}

This goes across the values passed in initial_state and predicts for them one at a time, as markovchainSequence() would expect. We can see that it works as expected:

x <- data.frame("AccountNo" = 1:5, 
                "CurrentState" = c("sunny", "sunny", "rain" , "cloudy" ,"rain"), 
                stringsAsFactors = FALSE)

set.seed(42)
x$Prediction3 <- predict(x$CurrentState,3)
x

  AccountNo CurrentState Prediction3
1         1        sunny       sunny
2         2        sunny       sunny
3         3         rain       sunny
4         4       cloudy        rain
5         5         rain       sunny