how to calculate cumsum with depreciation in a grouped dataframe? - r

Question

I tried to calculate the cumsum with a depreciation rate.
I have a grouped dataframe with a column number.
I want to add the number one by one with depreciation.
If the rate is 1, then the cumsum function in base r is good enough.
But if not, let's say the rate of 0.5 (means each number will multiply by 0.5 to add the next number), cumsum is not enough.
I tried to write my own function to work with dplyr, but it fails.
library(tidyverse)
# dataframe
id=sample(1:5,25,replace=TRUE)
num=rnorm(25)
df=data.frame(id,num)
# my custom function
depre=function(data){
rate=0.5
r=nrow(data)
sl=data$num
nl=data$num
for (i in 2:r){
sl[i]=sl[i-1]*rate+nl[i]
}
return(sl)
}
# work with one group
df %>% filter(id==1) %>% depre(.)
# failed to work with dplyr
df %>% group_by(id) %>% mutate(sl=depre(.))
I expect the first element of column s, should be the same as in column num.
But the following ones, should be depreciate by times 0.5 and add next num.
It works in one group, but failed in multi-grouped dataframe.
The error message is: "Error: Column sl must be length 6 (the group size) or one, not 25".
I have no idea. Could anyone have a clue?
Thanks

Answer 1

Your function would work if you pass vector to your function instead of dataframe
depre <- function(num){
rate = 0.5
r= length(num)
sl = num
nl = num
for (i in 2:r){
sl[i]=sl[i-1]*rate+nl[i]
}
return(sl)
}
and then apply it by group.
library(dplyr)
df %>% group_by(id) %>% mutate(sl = depre(num))

Answer 2

We can split by 'id' and use the OP's function without any changes
library(dplyr)
library(purrr)
df %>%
group_split(id, keep = FALSE) %>%
map_df(~ tibble(id = .$id, sl = depre(.)))
# id sl
# <int> <dbl>
# 1 1 1.07
# 2 1 -0.776
# 3 1 -0.518
# 4 1 0.628
# 5 1 0.601
# 6 1 1.10
# 7 2 -0.734
# 8 2 -0.583
# 9 2 -0.437
#10 2 -3.45
# … with 15 more rows
or an option would be accumulate from purrr which would be more compact
out <- df %>%
group_by(id) %>%
mutate(sl = accumulate(num, ~ .y + .x * 0.5))
out
# A tibble: 25 x 3
# Groups: id [5]
# id num sl
# <int> <dbl> <dbl>
# 1 3 -0.784 -0.784
# 2 2 -0.734 -0.734
# 3 2 -0.216 -0.583
# 4 3 -0.335 -0.727
# 5 5 -1.09 -1.09
# 6 4 -0.0854 -0.0854
# 7 1 1.07 1.07
# 8 2 -0.145 -0.437
# 9 3 -1.17 -1.53
#10 5 -0.819 -1.36
# … with 15 more rows
out %>%
filter(id == 1)
# A tibble: 6 x 3
# Groups: id [1]
# id num sl
# <int> <dbl> <dbl>
#1 1 1.07 1.07
#2 1 -1.31 -0.776
#3 1 -0.129 -0.518
#4 1 0.887 0.628
#5 1 0.287 0.601
#6 1 0.800 1.10
Issue in the OP's function is that the input is the whole dataset and during the process of getting the number of rows, it uses nrow(data), which would be the total number of rows. With group_by, the dplyr convention is n() - giving the number of rows. By doing the group_split, the input data.frame is split into subset of data.frames and the nrow of those will work for the created function