A running sum for daily data that resets when month turns - r

Question

I have a 2 column table (tibble), made up of a date object and a numeric variable. There is maximum one entry per day but not every day has an entry (ie date is a natural primary key). I am attempting to do a running sum of the numeric column along with dates but with the running sum resetting when the month turns (the data is sorted by ascending date). I have replicated what I want to get as a result below.
Date score monthly.running.sum
10/2/2019 7 7
10/9/2019 6 13
10/16/2019 12 25
10/23/2019 2 27
10/30/2019 13 40
11/6/2019 2 2
11/13/2019 4 6
11/20/2019 15 21
11/27/2019 16 37
12/4/2019 4 4
12/11/2019 24 28
12/18/2019 28 56
12/25/2019 8 64
1/1/2020 1 1
1/8/2020 15 16
1/15/2020 9 25
1/22/2020 8 33
It looks like the package "runner" is possibly suited to this but I don't really understand how to instruct it. I know I could use a join operation plus a group_by using dplyr to do this, but the data set is very very large and doing so would be wildly inefficient. i could also manually iterate through the list with a loop, but that also seems inelegant. last option i can think of is selecting out a unique vector of yearmon objects and then cutting the original list into many shorter lists and running a plain cumsum on it, but that also feels unoptimal. I am sure this is not the first time someone has to do this, and given how many tools there is in the tidyverse to do things, I think I just need help finding the right one. The reason I am looking for a tool instead of using one of the methods I described above (which would take less time than writing this post) is because this code needs to be very very readable by an audience that is less comfortable with code.

Answer 1

We can also use data.table
library(data.table)
setDT(df)[, Date := as.IDate(Date, "%m/%d/%Y")
][, monthly.running.sum := cumsum(score),format(Date, "%Y-%m")][]
# Date score monthly.running.sum
# 1: 2019-10-02 7 7
# 2: 2019-10-09 6 13
# 3: 2019-10-16 12 25
# 4: 2019-10-23 2 27
# 5: 2019-10-30 13 40
# 6: 2019-11-06 2 2
# 7: 2019-11-13 4 6
# 8: 2019-11-20 15 21
# 9: 2019-11-27 16 37
#10: 2019-12-04 4 4
#11: 2019-12-11 24 28
#12: 2019-12-18 28 56
#13: 2019-12-25 8 64
#14: 2020-01-01 1 1
#15: 2020-01-08 15 16
#16: 2020-01-15 9 25
#17: 2020-01-22 8 33
data
df <- structure(list(Date = c("10/2/2019", "10/9/2019", "10/16/2019",
"10/23/2019", "10/30/2019", "11/6/2019", "11/13/2019", "11/20/2019",
"11/27/2019", "12/4/2019", "12/11/2019", "12/18/2019", "12/25/2019",
"1/1/2020", "1/8/2020", "1/15/2020", "1/22/2020"), score = c(7L,
6L, 12L, 2L, 13L, 2L, 4L, 15L, 16L, 4L, 24L, 28L, 8L, 1L, 15L,
9L, 8L)), row.names = c(NA, -17L), class = "data.frame")

Answer 2

Using lubridate, you can extract month and year values from the date, group_by those values and them perform the cumulative sum as follow:
library(lubridate)
library(dplyr)
df %>% mutate(Month = month(mdy(Date)),
Year = year(mdy(Date))) %>%
group_by(Month, Year) %>%
mutate(SUM = cumsum(score))
# A tibble: 17 x 6
# Groups: Month, Year [4]
Date score monthly.running.sum Month Year SUM
<chr> <int> <int> <int> <int> <int>
1 10/2/2019 7 7 10 2019 7
2 10/9/2019 6 13 10 2019 13
3 10/16/2019 12 25 10 2019 25
4 10/23/2019 2 27 10 2019 27
5 10/30/2019 13 40 10 2019 40
6 11/6/2019 2 2 11 2019 2
7 11/13/2019 4 6 11 2019 6
8 11/20/2019 15 21 11 2019 21
9 11/27/2019 16 37 11 2019 37
10 12/4/2019 4 4 12 2019 4
11 12/11/2019 24 28 12 2019 28
12 12/18/2019 28 56 12 2019 56
13 12/25/2019 8 64 12 2019 64
14 1/1/2020 1 1 1 2020 1
15 1/8/2020 15 16 1 2020 16
16 1/15/2020 9 25 1 2020 25
17 1/22/2020 8 33 1 2020 33
An alternative will be to use floor_date function in order ot convert each date as the first day of each month and the calculate the cumulative sum:
library(lubridate)
library(dplyr)
df %>% mutate(Floor = floor_date(mdy(Date), unit = "month")) %>%
group_by(Floor) %>%
mutate(SUM = cumsum(score))
# A tibble: 17 x 5
# Groups: Floor [4]
Date score monthly.running.sum Floor SUM
<chr> <int> <int> <date> <int>
1 10/2/2019 7 7 2019-10-01 7
2 10/9/2019 6 13 2019-10-01 13
3 10/16/2019 12 25 2019-10-01 25
4 10/23/2019 2 27 2019-10-01 27
5 10/30/2019 13 40 2019-10-01 40
6 11/6/2019 2 2 2019-11-01 2
7 11/13/2019 4 6 2019-11-01 6
8 11/20/2019 15 21 2019-11-01 21
9 11/27/2019 16 37 2019-11-01 37
10 12/4/2019 4 4 2019-12-01 4
11 12/11/2019 24 28 2019-12-01 28
12 12/18/2019 28 56 2019-12-01 56
13 12/25/2019 8 64 2019-12-01 64
14 1/1/2020 1 1 2020-01-01 1
15 1/8/2020 15 16 2020-01-01 16
16 1/15/2020 9 25 2020-01-01 25
17 1/22/2020 8 33 2020-01-01 33

Answer 3

A base R alternative :
df$Date <- as.Date(df$Date, "%m/%d/%Y")
df$monthly.running.sum <- with(df, ave(score, format(Date, "%Y-%m"),FUN = cumsum))
df
# Date score monthly.running.sum
#1 2019-10-02 7 7
#2 2019-10-09 6 13
#3 2019-10-16 12 25
#4 2019-10-23 2 27
#5 2019-10-30 13 40
#6 2019-11-06 2 2
#7 2019-11-13 4 6
#8 2019-11-20 15 21
#9 2019-11-27 16 37
#10 2019-12-04 4 4
#11 2019-12-11 24 28
#12 2019-12-18 28 56
#13 2019-12-25 8 64
#14 2020-01-01 1 1
#15 2020-01-08 15 16
#16 2020-01-15 9 25
#17 2020-01-22 8 33

Answer 4

The yearmon class represents year/month objects so just convert the dates to yearmon and accumulate by them using this one-liner:
library(zoo)
transform(DF, run.sum = ave(score, as.yearmon(Date, "%m/%d/%Y"), FUN = cumsum))
giving:
Date score run.sum
1 10/2/2019 7 7
2 10/9/2019 6 13
3 10/16/2019 12 25
4 10/23/2019 2 27
5 10/30/2019 13 40
6 11/6/2019 2 2
7 11/13/2019 4 6
8 11/20/2019 15 21
9 11/27/2019 16 37
10 12/4/2019 4 4
11 12/11/2019 24 28
12 12/18/2019 28 56
13 12/25/2019 8 64
14 1/1/2020 1 1
15 1/8/2020 15 16
16 1/15/2020 9 25
17 1/22/2020 8 33