use rollapply for certain rows - r

Question

I wonder if it is possible to use rollapply() only for certain rows of a dataframe. I know the "by" argument can specify the every by-th time point at which I calculate FUN, but now I have a very specific vector of row indices to which I wish to apply the rollapply(). For example, I have the below dataframe:
df <- data.frame(x = (1:10), y = (11:20))
I know how to calculate the rolling mean for y column when the rolling width is 3.
library(zoo)
m <- rollapply(df$y, width = 3, FUN = mean, fill = NA, align = "right")
But what if I want the width-3-mean only for the 4th and 9th row? Is there something in "by" argument that I can manipulate? Or some other better methods (using apply to do rolling calculation maybe)?

Answer 1

Hopefully I am understanding your question correctly. I think you are asking how to perform a function on every 4th and 9th element in a sliding window? If yes, just restrict your function to the 4th and 9th element using x[4] and x[9]. Like this:
output <- rollapply(df, 9, function(x) (x[4] + x[9])/2), fill = NA, align = "right")
I also interpret your question as asking how to get the mean when the window contains the 4th or 9th row? This can be done by sub setting. The question you need to think about is where you want the 4th and 9th row to be located within your window. Do you want the 4th row to be at position x[1], x[2], or x[3] within your window? Depending on what is at the other positions will obviously effect your output. Say you dont know, and all three seem reasonable, you will need to write a function a that creates a list of dataframes containing the range of data you are interested in, and then use an apply function, or a for loop, to rollapply the mean function over each dataframe in the list. You can then all of these outputs into a dataframe to work with further. Like this:
# the rlist library has a function that allows us to add items to a list
# which will be handy later on
library(rlist)
library(zoo)
# your example data
df <- data.frame(x = (1:10), y = (11:20))
# a vector of your desired rows
desired_rows <- c(4,9)
# A for loop that generates a list of dataframes
# with your desired rows in the middle of each
for (i in desired_rows){
lower_bound <- i-2
upper_bound <- i+2
df_subset <- df[c(lower_bound:upper_bound), ]
if(exists("list_df_range")){
list_df_range <- list.append(list_df_range, df_subset)
}else{
list_df_range <- list(df_subset)
}
}
# a second for loop that applies your rollapply function to each
# data frame in the list and then
# returns a dataframe of the final results
# with each column named after the originating row
for (n in list_df_range){
m <- rollapply(n$y, width = 3, FUN = mean, fill = NA, align = "right")
if(exists("final_out")){
final_out <- cbind(final_out, m)
}else{
final_out <- data.frame(m)
}
}
names(final_out) <- desired_rows

Answer 2

Based on the comment below the question by the poster it seems that what is wanted is to take the mean of each rolling window of width 3 excluding the middle element in each window and only keeping the 4th and 9th elements so
cc <- c(4, 9)
rollapply(df$y, list(c(-2, 0)), mean, fill = NA)[cc]
## [1] 13 18
or
rollapplyr(df$y, 3, function(x) mean(x[-2]), fill = NA)[cc]
## [1] 13 18
or
sapply(cc, function(ix) mean(df$y[seq(to = ix, by = 2, length = 2)]))
## [1] 13 18
or
(df$y[cc - 2] + df$y[cc]) / 2
## [1] 13 18