improvement on tapply (shifting groups of vectors)

Question

The order of the return object from tapply() is ambiguous, so I've started to worry about this bit of code:

#d <- data.frame(value = c(1,2,3,5), 
#               source = c("a","a","b","b"))
d$value <- unlist(tapply(d$value, d$source, function(v) v-v[1]))

The idea is to split value into groups, then and then shift each group of its element so that the group starts off at 0.

What's a better way to do this? I have several ideas, but I can't think of anything that

• doesn't make use of third-party libraries
• is still a one-liner
• avoids an R loop
• does not involve sorting/merging the original data frame.

Answer 1

A little-known trick in base R is the split(x, y) <- lapply(split(x, y), f) paradigm, so the following one-liner meets all your requirements:

d <- data.frame(value = c(1,2,3,5), 
                source = c("a","a","b","b"))

split(d$value, d$source) <- lapply(split(d$value, d$source), \(x) x - x[1])

Resulting in:

d
#>   value source
#> 1     0      a
#> 2     1      a
#> 3     0      b
#> 4     2      b

Whatever the ordering of source

^{Created on 2023-06-06 with reprex v2.0.2}

Answer 2

Hope ave makes sense

> transform(d,value = ave(value,source, FUN = \(x) x-x[1]))
  value source
1     0      a
2     1      a
3     0      b
4     2      b