How to average adjacent columns (non-overlapping) in R using dplyr function? - r

Question

This is a sample of my dataset.
library(tidyr)
library(dplyr)
resource <- c("good","good","bad","bad","good","good","bad","bad","good","good","bad","bad","good","good","bad","bad")
fertilizer <- c("none", "nitrogen","none","nitrogen","none", "nitrogen","none","nitrogen","none", "nitrogen","none","nitrogen","none", "nitrogen","none","nitrogen")
t1 <- sample(1:20, 16)
t2 <- sample(1:20, 16)
t3 <- sample(1:20, 16)
t4 <- sample(1:20, 16)
t5 <- sample(1:20, 16)
t6 <- sample(10:100, 16)
t7 <- sample(10:100, 16)
t8 <- sample(10:100, 16)
t9 <- sample(10:100, 16)
t10 <- sample(10:100, 16)
replicates <- c(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16)
data <- data.frame(resource, fertilizer,replicates, t1,t2,t3,t4,t5,t6,t7,t8,t9,t10)
data$resource <- as.factor(data$resource)
data$fertilizer <- as.factor(data$fertilizer)
Where t0,t1,t2..etc are time points. I need to average adjacent time points (non-overlapping) across eg. (t1,t2), (t3,t4)..and the new column headings need to have the average of the times, so that the columns read as t1.5,t3.5,...etc.
Thus in the end I need to have only 5 columns reading t1.5, t3.5,t5.5, t7.5,t9.5
Is there anyway this can be achieved using dplyr function, or any other function in R?

Answer 1

Edited for OP's modified request:
If you put everything in a tidy format, you can take advantage of the lag/lead functions to average adjacent rows.
library(stringr)
library(forcats)
data %>%
gather(key = time, value = value, -replicates, -resource, -fertilizer) %>%
mutate(index = as.integer(str_extract(time, "[0-9]+"))) %>%
arrange(replicates, index) %>%
group_by(resource, fertilizer, replicates) %>%
mutate(mid_value = (value + lead(value))/2,
mid_index = (index + lead(index))/2,
mid_time = str_c("t",mid_index)) %>%
ungroup %>%
filter(!is.na(mid_value), index %% 2 == 1) %>%
select(replicates, resource, fertilizer, matches("mid")) %>%
rename(value = mid_value, time = mid_time, index = mid_index) %>%
arrange(index) %>%
mutate(time = as_factor(time)) %>%
select(-index) %>%
spread(key = time, value = value) %>%
arrange(replicates)

Answer 2

Solution using only base R: You need to somehow find the columns you want to calculate the average for. You can do this by searching the column names for the t + "somenumber" pattern. After that, create a sequence of sequences, corresponding to the column numbers of df you want to calculate the mean for.
relevant_cols <- grep("[0-9]{1,2}", names(df))
start <- min(relevant_cols)
end <- max(relevant_cols)
cols <- split(start:end, rep(1:5, each=2))
If you look at cols, you'll see that it is list of five, each element resembling a combination of columns you want to average. This smells like a use-case for sapply():
newdf <- sapply(cols, function(x) rowMeans(df[x]) )
colnames(newdf) <- paste0("t", seq(1, diff(range(relevant_cols)), 2) + 0.5)
Edit: I seem to have misunderstood what you want to maintain and what not. You can just cbind() (parts of) the old df to newdf:
cbind(df, newdf)
cbind(df[, -relevant_cols], newdf) # This is what you want. I think..

Answer 3

Here ya go:
transmute(data,
t1.5 = (t1 + t2) / 2,
t3.5 = (t3 + t4) / 2,
t5.5 = (t5 + t6) / 2,
t7.5 = (t7 + t8) / 2,
t9.5 = (t9 + t10) / 2)