CODEWITHSUNDEEP
r
Pk.yd
Pk.yd

Reputation: 311

R loops+predict()

Hi I am a beginner with R (beginner programmer in general) and the help documents are absolutely killing me.

Suppose I have a matrix [a,b,c,d] I complete 2 regression of some kind a~b+c+d. My goal is to do a predict() for the variable "a" in test data set but c is full of NAs. How do I replace the NAs in c using the model I have created?

If it helps this is the kind of loop I would do in Octave,

 for i:length(c)
    if c(i)=NA  
    c(i)=some_function(b,d);<---- I tried to bold this but it came out wrong
 end

Thanks

Upvotes: 0

Views: 845

Answers (2)

Richie Cotton
Richie Cotton

Reputation: 121127

It's even easier than Seb suggests.

c[is.na(c)] <- mean(c, na.rm = TRUE)

Here, the mean function returns a single number (namely the mean of all the values in c that weren't NA). The assignment operator <- then assigns this number to every element in c where is.na returns TRUE.

As an alternative, try passing the argument na.action = na.omit to the predict function.

The direct translation of your Octave script is something like

for(i in seq_along(c))
{
  if(is.na(c[i]))
  {  
    c(i) <- some_function(b[i], d[i])
  }
}

Note however that in R, just as in Octave, loops are usually inferior to operating directly on vectors.

Upvotes: 1

Seb
Seb

Reputation: 5507

do you mean something like

c <- ifelse(is.na(c), mean(c, na.rm=TRUE), c)

you may want to check the help files ?ifelse and ?is.na.

Upvotes: 0

Related Questions