How to cumsum the elements of a vector under certain condition in R?

user_15349622 0
Viewed 187

My objective is to do a cumulative sum of the elements of a vector and assign the result to each element. But when certain condition is reached, then reset the cumulative sum.

For example:

vector_A <- c(1, 1, -1, -1, -1, 1, -1, -1, 1, -1)

Now, suppose that the condition to reset the cumulative sum is that the next element has a different sign.

Then the desired output is:

vector_B <- c(1, 2, -1, -2, -3, 1, -1, -2, 1, -1)

How can I achieve this?

cumsum conditional-statements if-statement vector r
4 Answers

You can use a custom function instead of cumsum and accumulate results using e.g. purrr::accumulate:

library(purrr)
vector_A <- c(1, 1, -1, -1, -1, 1, -1, -1, 1, -1)

purrr::accumulate(vector_A, function(a,b) {
  if (sign(a) == sign(b))
    a+b
  else
    b
  })

[1]  1  2 -1 -2 -3  1 -1 -2  1 -1

or if you want to avoid any branch:

purrr::accumulate(vector_A, function(a,b) { b + a*(sign(a) == sign(b))})

[1]  1  2 -1 -2 -3  1 -1 -2  1 -1
user_32075090

A base R option with Reduce

> Reduce(function(x, y) ifelse(x * y > 0, x + y, y), vector_A, accumulate = TRUE)
 [1]  1  2 -1 -2 -3  1 -1 -2  1 -1

or using ave + cumsum

> ave(vector_A, cumsum(c(1, diff(sign(vector_A)) != 0)), FUN = cumsum)
 [1]  1  2 -1 -2 -3  1 -1 -2  1 -1
user_121587570

Using ave:

ave(vector_A, data.table::rleid(sign(A)), FUN = cumsum)
#  [1]  1  2 -1 -2 -3  1 -1 -2  1 -1

A formula version of accumulate:

purrr::accumulate(vector_A, ~ ifelse(sign(.x) == sign(.y), .x + .y, .y))
#  [1]  1  2 -1 -2 -3  1 -1 -2  1 -1
user_134606020

The approach that comes to mind is to find the runs (rle()) defined by the condition (sign()) in the data, apply cumsum() on each run separately (tapply()), and the concatenate back into a vector (unlist()). Something like this:

vector_A <- c(1, 1, -1, -1, -1, 1, -1, -1, 1, -1)

run_length <- rle(sign(vector_A))$lengths
run_id <- rep(seq_along(run_length), run_length)

unlist(tapply(vector_A, run_id, cumsum), use.names = FALSE)
#>  [1]  1  2 -1 -2 -3  1 -1 -2  1 -1

Wrapping the process up a bit, I’d maybe put finding the grouping factor (run index) in a function? And then the grouped summary will need to be done using existing tools, like tapply() above, or a creative ave(), or in the context of data frames, a group_by() and summarise() with dplyr.

run_index <- function(x) {
  with(rle(x), rep(seq_along(lengths), lengths))
}

ave(vector_A, run_index(sign(vector_A)), FUN = cumsum)
#>  [1]  1  2 -1 -2 -3  1 -1 -2  1 -1
user_45506950
Related