data.table difference between rows of non-consecutive records of same group

Question

I need to calculate the difference between non-consecutive records of a variable, grouped by another. That is, I want to take the last value of the variable in a run and subtract it from the first value in the next run (if there is any).

I know I can use rleid along with shift to calculate differences in consecutive rows, but this time I need to get rid of those.

Example data

dput(iris)

structure(list(Sepal.Length = c(4.4, 6.3, 4.6, 5.8, 6.4, 6.5, 
4.9, 5.4, 6.4, 6.7), Sepal.Width = c(3, 2.8, 3.1, 2.7, 2.7, 3, 
3.6, 3.9, 2.8, 3.1), Petal.Length = c(1.3, 5.1, 1.5, 4.1, 5.3, 
5.5, 1.4, 1.7, 5.6, 4.7), Petal.Width = c(0.2, 1.5, 0.2, 1, 1.9, 
1.8, 0.1, 0.4, 2.1, 1.5), Species = c("setosa", "virginica", 
"setosa", "versicolor", "virginica", "virginica", "setosa", "setosa", 
"virginica", "versicolor")), .Names = c("Sepal.Length", "Sepal.Width", 
"Petal.Length", "Petal.Width", "Species"), row.names = c(NA, 
-10L), class = c("data.table", "data.frame"))

library(data.table)
setDT(iris, key = "Sepal.Width")

I thnik something like

 iris[, diff(Petal.Width), by = .(Species, !rleid(Species))]

(of course this doesn't work!) is what I need, but can't think of anything to achieve it.

Expected result (diffing Petal.Width):

      Species   V1
1: versicolor  0.5
2:  virginica -0.3
3:     setosa  0.0
4:     setosa -0.1

(I achieved it doing iris[, diff(Petal.Width), by = .(Species)] and then hand-picking .Last.Value[, c(1, 4, 5, 6)])

Answer 1

Well, there's

iris[, .(first(Petal.Width), last(Petal.Width)), by=.(Species, rleid(Species))][, 
  tail(V1 - shift(V2), -1), by=Species]

      Species   V1
1: versicolor  0.5
2:  virginica -0.3
3:     setosa  0.0
4:     setosa -0.1

Or...

iris[, Petal.Width[c(1L, .N)], by=.(Species, rleid(Species))][, {
  v = V1[-c(1L, .N)]
  v[c(TRUE,FALSE)] - v[c(FALSE,TRUE)]
}, by=Species]

      Species   V1
1: versicolor -0.5
2:  virginica  0.3
3:     setosa  0.0
4:     setosa  0.1

Answer 2

After trying several things, I came with this hacky solution. I guess there sould be something "cleaner":

iris[, .(Species, Petal.Width, rl= rleid(Species))][, .(pd= ifelse(diff(rl)>0, diff(Petal.Width), NA)), by = Species][!is.na(pd),]

If there's a function to achieve this in a cleanner way, I appreciate the pointer.