Rolling means and applying means at beginning of a series of data

Question

I want to do a rolling mean of the the previous 4 values in a dataset. However, for the beginning, since there is not 4 values, I want to do the rolling mean of 1/2/3 observations. How do I do this?

 library(zoo)
 df= data.frame(a=c(1,2,3,4,5))
 df$answer = rollapply(df$a, 4,mean)
 #help

For example, row 1 would have a value of 1, row 2 would have a value of (1+2)/2=1.5, row 3 would have a value of 6/3=2.

I want to do rolling means of 4 periods but in periods with fewer months, I want to do the mean of the maximum periods allowed.

Answer 1

Use right aligment with partial=TRUE, i.e. rollapplyr(..., partial=TRUE) or rollapply(..., align = "right", partial=TRUE). Here we use rollapplyr:

rollapplyr(df$a, 4, mean, partial = TRUE)

Answer 2

I think it can be simply done with a simple function such as the following (as an alternative solution):

rollapply2 <- function(myvec, width, fun){
  #the first values up to width
  firstvalues  <- cumsum(myvec[1:(width-1)])/(1:(width-1))
  #the rest of the values as normal
  normalvalues <- rollapply(myvec, width, fun)
  #return them all
  c(firstvalues,normalvalues)
}

Output:

> rollapply2(df$a, 4, mean)
[1] 1.0 1.5 2.0 2.5 3.5

Answer 3

You can also try without package:

sapply(seq_along(df$a), function(u) mean(df$a[max(u-3,0):u]))
#[1] 1.0 1.5 2.0 2.5 3.5

Or a vectorized solution - without loop - base R:

with(df, (cumsum(a) - c(rep(0,4),head(a,-4)))/pmin(seq_along(a),4))
#[1] 1.0 1.5 2.0 2.5 3.5

Answer 4

What about adding extra NAs ?

rollapply(c(rep(NA, 3),df$a), 4, FUN = mean, align = "right", na.rm = TRUE)