In R, how can I calculate cumsum for a defined time period prior to the row being calculate? Prefer dplyr if possible.
For example, if the period was 10 days, then the function would achieve cum_rolling10:
date value cumsum cum_rolling10
1/01/2000 9 9 9
2/01/2000 1 10 10
5/01/2000 9 19 19
6/01/2000 3 22 22
7/01/2000 4 26 26
8/01/2000 3 29 29
13/01/2000 10 39 29
14/01/2000 9 48 38
18/01/2000 2 50 21
19/01/2000 9 59 30
21/01/2000 8 67 38
25/01/2000 5 72 24
26/01/2000 1 73 25
30/01/2000 6 79 20
31/01/2000 6 85 18
A solution using dplyr, tidyr, lubridate, and zoo.
library(dplyr)
library(tidyr)
library(lubridate)
library(zoo)
dt2 <- dt %>%
mutate(date = dmy(date)) %>%
mutate(cumsum = cumsum(value)) %>%
complete(date = full_seq(date, period = 1), fill = list(value = 0)) %>%
mutate(cum_rolling10 = rollapplyr(value, width = 10, FUN = sum, partial = TRUE)) %>%
drop_na(cumsum)
dt2
# A tibble: 15 x 4
date value cumsum cum_rolling10
<date> <dbl> <int> <dbl>
1 2000-01-01 9 9 9
2 2000-01-02 1 10 10
3 2000-01-05 9 19 19
4 2000-01-06 3 22 22
5 2000-01-07 4 26 26
6 2000-01-08 3 29 29
7 2000-01-13 10 39 29
8 2000-01-14 9 48 38
9 2000-01-18 2 50 21
10 2000-01-19 9 59 30
11 2000-01-21 8 67 38
12 2000-01-25 5 72 24
13 2000-01-26 1 73 25
14 2000-01-30 6 79 20
15 2000-01-31 6 85 18
DATA
dt <- structure(list(date = c("1/01/2000", "2/01/2000", "5/01/2000",
"6/01/2000", "7/01/2000", "8/01/2000", "13/01/2000", "14/01/2000",
"18/01/2000", "19/01/2000", "21/01/2000", "25/01/2000", "26/01/2000",
"30/01/2000", "31/01/2000"), value = c(9L, 1L, 9L, 3L, 4L, 3L,
10L, 9L, 2L, 9L, 8L, 5L, 1L, 6L, 6L)), .Names = c("date", "value"
), row.names = c(NA, -15L), class = "data.frame")
this solution will avoid memory overhead, and migrate to sparklyr will be easy.
lag = 7
dt %>%
mutate(date = dmy(date)) %>%
mutate(order = datediff(date,min(date)) %>%
arrange(desc(order)) %>%
mutate(n_order = lag(order + lag,1L,default = 0)) %>%
mutate(b_order = ifelse(order - n_order >= 0,order,-1)) %>%
mutate(m_order = cummax(b_order)) %>%
group_by(m_order) %>%
mutate(rolling_value = cumsum(value))
来源：https://stackoverflow.com/questions/46396417/r-cumulative-sum-over-rolling-date-range