efficiently calculate annual cumulative sum

Question

I have a dataset with quarterly transactions. PERIOD represents the quarter of the transaction and INCREM represents the incremental amounts.

tbl <- data.frame(PERIOD = c(2,3,6,10,11),
                  INCREM = c(10,50,-30,-10,-20))

I want to get annual cumulative sums (so the cumulative sum at periods 4, 8, 12).

library(dplyr)
library(tidyr)

tbl %>%
  mutate(CUMSUM = cumsum(INCREM)) %>%
  select(-INCREM) %>%
  mutate(PERIOD = factor(PERIOD, 1:12)) %>%
  complete(PERIOD) %>%
  fill(CUMSUM) %>%
  mutate(PERIOD = as.numeric(PERIOD)) %>%
  filter(PERIOD %% 4 == 0)

Result:

  PERIOD CUMSUM
1      4     60
2      8     30
3     12      0

This works, but it's not very efficient. The original dataset is 5 rows and the final dataset is 3 rows, but in the middle of the dplyr chain (after fill() ) the dataset is 12 rows.

Is there a more efficient way to get the annual cumulative sums?

Also, my actual data is coming from a database query. Do you think it would be better for me to take care of this cumulative summing in the SQL query before manipulating in R?

Answer 1

cut is definitely the way to go. You can also just calculate the cumulative sum and then keep the final rows of the period. This avoids the aggregate step.

tbl$prd <- cut(tbl$PERIOD, c(1,4,8,Inf), labels=c(4,8,12))
tbl$cumsum <- cumsum(tbl$INCREM)
tbl[!duplicated(tbl$prd, fromLast=TRUE),c("prd","cumsum")]
#   prd cumsum
# 2   4     60
# 3   8     30
# 5  12      0

Answer 2

As @thelatemail suggested you can use cut to create groups, then sum values in each group and finally cumsum over all the values.

library(dplyr)
tbl %>%
  group_by(quarter = cut(PERIOD, c(1,4,8,Inf), labels=c(4,8,12))) %>%
  summarise(CUMSUM = sum(INCREM)) %>%
  ungroup() %>%
  mutate(CUMSUM = cumsum(CUMSUM))

#  quarter CUMSUM
#  <fct>    <dbl>
#1   4       60
#2   8       30
#3  12        0

---

Using same logic an overly complicated base R approach to fit in one line is

transform(aggregate(INCREM~PERIOD, 
  transform(tbl, PERIOD = cut(PERIOD, c(1,4,8,Inf), labels=c(4,8,12))), sum), 
    INCREM = cumsum(INCREM))


#  PERIOD INCREM
#1      4     60
#2      8     30
#3     12      0

which actually means

tbl$PERIOD <- cut(tbl$PERIOD, c(1,4,8,Inf), labels=c(4,8,12))
tbl1 <- aggregate(INCREM~PERIOD, tbl, sum)
tbl1$INCREM <- cumsum(tbl1$INCREM)