Determine windows/epochs in time series and calculate average - r

Question

I'm dealing with time series data from neurophysiological recordings which typically have 'markers' that mark the beginning of an event (e.g., stimulus being presented on the screen). I'm trying to subset particular windows/epochs based on certain markers, and then average those separate windows/epochs.
To illustrate this, below is really simple example (my actual datasets have millions of data points, so it'd be nice to have efficient solutions).
df <- data.frame(value = c(1:10, 101:110), #time series data
marker = c(NA, NA, 'start', NA, NA, NA, NA, 'end', NA, NA, #event markers
NA, NA, 'start', NA, NA, NA, NA, 'end', NA, NA))
start <- which(df$marker == "start") #indices 3 and 13 are the 'start' markers
end <- which(df$marker == 'end') #indices 8 and 18 are the 'end markers'
window1 <- df$value[start[1]:end[1]] #first window (indices 3 to 8)
window2 <- df$value[start[2]:end[2]] #second window (indices 13 to 18)
averageWindow <- (window1 + window2) / 2 #average of the two windows
Is this the most efficient way to go about doing this (I have nearly 1000 windows in my actual data and about 1 million rows)?

Answer 1

I am not sure if you want an average based on all windows or an average for each window. So I decided to produce both results. Using your start and end, I subsetted the data with lapply(). By this time, I removed irrelevant data. Then, I combined data frames in the list with rbindlist() and assigned ID to a new column. The final process was to get an average.
library(data.table)
start <- which(df$marker == "start") #indices 3 and 13 are the 'start' markers
end <- which(df$marker == 'end') #indices 8 and 18 are the 'end markers'
rbindlist(lapply(1:length(start), function(x){
df[start[x]:end[x], ]}), idcol = TRUE) -> temp
# An overall average
temp[, list(average = sum(value) / uniqueN(.id))]
# average
#1: 333
# An average for each window
temp[, list(average = sum(value) / .N), by = .id]
# .id average
#1: 1 5.5
#2: 2 105.5
Replying to the OP's message, I came up with the following code. I created ID for each of the 6 points and calculated an average for each point.
temp[, index := 1:.N, by = .id][,
list(average = sum(value) / .N), by = index]
# index average
#1: 1 53
#2: 2 54
#3: 3 55
#4: 4 56
#5: 5 57
#6: 6 58
DATA
df <- data.frame(value = c(1:10, 101:110), #time series data
marker = c(NA, NA, 'start', NA, NA, NA, NA, 'end', NA, NA, #event markers
NA, NA, 'start', NA, NA, NA, NA, 'end', NA, NA),
stringsAsFactors = FALSE)