我有一个随时间推移包含事件和子事件的数据表，我感兴趣的是创建两列：（1）事件发生后5年内是否发生了事件的累积滚动总和； （2）对自事件日期起5年内发生的子事件（包括事件）数量的计数。下面是带有代码的示例：
dt = data.table(id=c(rep(52749, 14), rep(46760, 15)),
date=c("2007-01-30","2007-03-15","2007-11-27",
"2007-11-29","2008-10-09","2009-04-02",
"2011-01-06","2011-07-26","2012-01-25",
"2015-01-12","2016-09-13","2017-03-21",
"2017-08-29","2017-10-10","2008-01-01",
"2010-07-19","2011-01-14","2011-08-02",
"2011-08-02","2012-02-01","2012-02-01",
"2015-04-28","2015-10-19","2016-05-16",
"2016-12-22","2016-12-23","2017-05-16",
"2017-11-15","2018-02-22"),
idx=c(seq_len(14), seq_len(15)),
count=c(rep(14,14),rep(15,15)),
event=c(1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 0, 1,
1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 0, 0, 1, 0))
产生的结果如下：
id date idx count event
52749 2007-01-30 1 14 1
52749 2007-03-15 2 14 0
52749 2007-11-27 3 14 1
52749 2007-11-29 4 14 0
52749 2008-10-09 5 14 1
52749 2009-04-02 6 14 0
52749 2011-01-06 7 14 1
52749 2011-07-26 8 14 1
52749 2012-01-25 9 14 0
52749 2015-01-12 10 14 1
52749 2016-09-13 11 14 1
52749 2017-03-21 12 14 1
52749 2017-08-29 13 14 0
52749 2017-10-10 14 14 0
46760 2008-01-01 1 15 1
46760 2010-07-19 2 15 1
46760 2011-01-14 3 15 0
46760 2011-08-02 4 15 1
46760 2011-08-02 5 15 0
46760 2012-02-01 6 15 1
46760 2012-02-01 7 15 0
46760 2015-04-28 8 15 1
46760 2015-10-19 9 15 0
46760 2016-05-16 10 15 1
46760 2016-12-22 11 15 1
46760 2016-12-23 12 15 0
46760 2017-05-16 13 15 0
46760 2017-11-15 14 15 1
46760 2018-02-22 15 15 0
我基本上需要以下内容：
id date idx count event num_event_5yr_fu num_subevents
52749 2007-01-30 1 14 1 4 8
52749 2007-03-15 2 14 0 NA NA
52749 2007-11-27 3 14 1 3 6
52749 2007-11-29 4 14 0 NA NA
52749 2008-10-09 5 14 1 2 4
52749 2009-04-02 6 14 0 NA NA
52749 2011-01-06 7 14 1 2 3
52749 2011-07-26 8 14 1 1 2
52749 2012-01-25 9 14 0 NA NA
52749 2015-01-12 10 14 1 2 4
52749 2016-09-13 11 14 1 1 3
52749 2017-03-21 12 14 1 0 2
52749 2017-08-29 13 14 0 NA NA
52749 2017-10-10 14 14 0 NA NA
46760 2008-01-01 1 15 1 3 6
46760 2010-07-19 2 15 1 3 6
46760 2011-01-14 3 15 0 NA NA
46760 2011-08-02 4 15 1 3 6
46760 2011-08-02 5 15 0 NA NA
46760 2012-02-01 6 15 1 3 6
46760 2012-02-01 7 15 0 NA NA
46760 2015-04-28 8 15 1 3 7
46760 2015-10-19 9 15 0 NA NA
46760 2016-05-16 10 15 1 2 5
46760 2016-12-22 11 15 1 1 4
46760 2016-12-23 12 15 0 NA NA
46760 2017-05-16 13 15 0 NA NA
46760 2017-11-15 14 15 1 0 1
46760 2018-02-22 15 15 0 NA NA
其中num_event_5yr_fu在计算从事件日期（不包括事件日期）起5年内一个事件发生的次数（或该事件的累积总和），而num_subevents为计算从事件日期（不包括事件日期）起的5年内的记录数。
我已经待了很长时间了，被困住了，我真的很感谢在此方面的一些投入。谢谢。
答案 0 :(得分：1)
以下是使用非等额联接的data.table方法：
library(lubridate)
dt[, date := as.Date(date)]
dt[, end_date := date]
year(dt$end_date) <- year(dt$end_date) + 5
dt[, rowid := .I]
event_count = dt[dt, on = .(date < date , end_date >= date, id),
allow.cartesian=TRUE][!is.na(rowid) & event == 1,
.(events = sum(i.event), num_subevents = .N),
by = .(rowid, id)]
dt[event_count, on = .(rowid, id), `:=`(num_event_5yr_fu = i.events,
num_subevents = i.num_subevents)]
dt[, c("end_date", "rowid") := NULL]
dt
# id date idx count event num_event_5yr_fu num_subevents
# 1: 52749 2007-01-30 1 14 1 4 8
# 2: 52749 2007-03-15 2 14 0 NA NA
# 3: 52749 2007-11-27 3 14 1 3 6
# 4: 52749 2007-11-29 4 14 0 NA NA
# 5: 52749 2008-10-09 5 14 1 2 4
# 6: 52749 2009-04-02 6 14 0 NA NA
# 7: 52749 2011-01-06 7 14 1 2 3
# 8: 52749 2011-07-26 8 14 1 1 2
# 9: 52749 2012-01-25 9 14 0 NA NA
# 10: 52749 2015-01-12 10 14 1 2 4
# 11: 52749 2016-09-13 11 14 1 1 3
# 12: 52749 2017-03-21 12 14 1 0 2
# 13: 52749 2017-08-29 13 14 0 NA NA
# 14: 52749 2017-10-10 14 14 0 NA NA
# 15: 46760 2008-01-01 1 15 1 3 6
# 16: 46760 2010-07-19 2 15 1 3 6
# 17: 46760 2011-01-14 3 15 0 NA NA
# 18: 46760 2011-08-02 4 15 1 3 5
# 19: 46760 2011-08-02 5 15 0 NA NA
# 20: 46760 2012-02-01 6 15 1 3 5
# 21: 46760 2012-02-01 7 15 0 NA NA
# 22: 46760 2015-04-28 8 15 1 3 7
# 23: 46760 2015-10-19 9 15 0 NA NA
# 24: 46760 2016-05-16 10 15 1 2 5
# 25: 46760 2016-12-22 11 15 1 1 4
# 26: 46760 2016-12-23 12 15 0 NA NA
# 27: 46760 2017-05-16 13 15 0 NA NA
# 28: 46760 2017-11-15 14 15 1 0 1
# 29: 46760 2018-02-22 15 15 0 NA NA
答案 1 :(得分：1)
另一个选择：
library(data.table)
library(lubridate)
dt[, date := as.Date(date)][
, num_event_5yr_fu := sapply(date,
function(x) sum(event[between(date, x + 1, x + years(5))])), by = id
][, num_subevents := sapply(date,
function(x) length(event[between(date, x + 1, x + years(5))])), by = id
][event == 0, `:=` (num_event_5yr_fu = NA, num_subevents = NA)]
输出：
id date idx count event num_event_5yr_fu num_subevents
1: 52749 2007-01-30 1 14 1 4 8
2: 52749 2007-03-15 2 14 0 NA NA
3: 52749 2007-11-27 3 14 1 3 6
4: 52749 2007-11-29 4 14 0 NA NA
5: 52749 2008-10-09 5 14 1 2 4
6: 52749 2009-04-02 6 14 0 NA NA
7: 52749 2011-01-06 7 14 1 2 3
8: 52749 2011-07-26 8 14 1 1 2
9: 52749 2012-01-25 9 14 0 NA NA
10: 52749 2015-01-12 10 14 1 2 4
11: 52749 2016-09-13 11 14 1 1 3
12: 52749 2017-03-21 12 14 1 0 2
13: 52749 2017-08-29 13 14 0 NA NA
14: 52749 2017-10-10 14 14 0 NA NA
15: 46760 2008-01-01 1 15 1 3 6
16: 46760 2010-07-19 2 15 1 3 6
17: 46760 2011-01-14 3 15 0 NA NA
18: 46760 2011-08-02 4 15 1 3 5
19: 46760 2011-08-02 5 15 0 NA NA
20: 46760 2012-02-01 6 15 1 3 5
21: 46760 2012-02-01 7 15 0 NA NA
22: 46760 2015-04-28 8 15 1 3 7
23: 46760 2015-10-19 9 15 0 NA NA
24: 46760 2016-05-16 10 15 1 2 5
25: 46760 2016-12-22 11 15 1 1 4
26: 46760 2016-12-23 12 15 0 NA NA
27: 46760 2017-05-16 13 15 0 NA NA
28: 46760 2017-11-15 14 15 1 0 1
29: 46760 2018-02-22 15 15 0 NA NA
答案 2 :(得分：0)
OP的规格与OP的预期结果之间存在偏差。
OP已指定 num_event_5yr_fu对从事件日期（不包括事件日期）起的5年内事件发生的次数（或此事件的累计总和）进行计数，并且num_subevents正在计算自活动日期（不包括活动日期）起5年内的记录数。
但是，在OP的预期结果中，num_subevents正在计算从事件日期（不包括事件 行）起5年内的记录的数量（=记录？）。
因此，提供了两种解决方案，涵盖了两种解释。
此方法再现了OP的预期结果（与arg0naut和docendo discimus的答案相反，它们实现了OP的描述）。
此方法以非等额联接方式聚集和更新。它在联接中包括事件日期，但会更正总计以减少一个事件。
library(data.table)
new_cols <- c("num_event_5yr_fu", "num_subevents")
result <- dt[
, date := as.Date(date)][
.(id = id, start = date, end = date + lubridate::years(5)),
on = .(id, date >= start, date <= end),
new_cols := .(sum(event) - 1, .N - 1L), by = .EACHI][
event == 0, new_cols := NA][]
result
id date idx count event num_event_5yr_fu num_subevents 1: 52749 2007-01-30 1 14 1 4 8 2: 52749 2007-03-15 2 14 0 NA NA 3: 52749 2007-11-27 3 14 1 3 6 4: 52749 2007-11-29 4 14 0 NA NA 5: 52749 2008-10-09 5 14 1 2 4 6: 52749 2009-04-02 6 14 0 NA NA 7: 52749 2011-01-06 7 14 1 2 3 8: 52749 2011-07-26 8 14 1 1 2 9: 52749 2012-01-25 9 14 0 NA NA 10: 52749 2015-01-12 10 14 1 2 4 11: 52749 2016-09-13 11 14 1 1 3 12: 52749 2017-03-21 12 14 1 0 2 13: 52749 2017-08-29 13 14 0 NA NA 14: 52749 2017-10-10 14 14 0 NA NA 15: 46760 2008-01-01 1 15 1 3 6 16: 46760 2010-07-19 2 15 1 3 6 17: 46760 2011-01-14 3 15 0 NA NA 18: 46760 2011-08-02 4 15 1 3 6 19: 46760 2011-08-02 5 15 0 NA NA 20: 46760 2012-02-01 6 15 1 3 6 21: 46760 2012-02-01 7 15 0 NA NA 22: 46760 2015-04-28 8 15 1 3 7 23: 46760 2015-10-19 9 15 0 NA NA 24: 46760 2016-05-16 10 15 1 2 5 25: 46760 2016-12-22 11 15 1 1 4 26: 46760 2016-12-23 12 15 0 NA NA 27: 46760 2017-05-16 13 15 0 NA NA 28: 46760 2017-11-15 14 15 1 0 1 29: 46760 2018-02-22 15 15 0 NA NA id date idx count event num_event_5yr_fu num_subevents
请注意，第18至20行（2011-08-02至2012-02-01之间的id == 46760和date）符合OP的预期结果。
可以通过
进行验证all.equal(result, expected, check.attributes = FALSE)
[1] TRUE
此处，仅记录日期大于事件日期的记录。
library(data.table)
tmp <- dt[, date := as.Date(date)][
dt[event == 1, .(id, start = date, end = date + lubridate::years(5))],
on = .(id, date > start, date <= end),
.(event = 1, sum(event), .N), by = .EACHI]
result <- dt[tmp, on = .(id, event, date),
c("num_event_5yr_fu", "num_subevents") := .(V2, N)][]
result
id date idx count event num_event_5yr_fu num_subevents 1: 52749 2007-01-30 1 14 1 4 8 2: 52749 2007-03-15 2 14 0 NA NA 3: 52749 2007-11-27 3 14 1 3 6 4: 52749 2007-11-29 4 14 0 NA NA 5: 52749 2008-10-09 5 14 1 2 4 6: 52749 2009-04-02 6 14 0 NA NA 7: 52749 2011-01-06 7 14 1 2 3 8: 52749 2011-07-26 8 14 1 1 2 9: 52749 2012-01-25 9 14 0 NA NA 10: 52749 2015-01-12 10 14 1 2 4 11: 52749 2016-09-13 11 14 1 1 3 12: 52749 2017-03-21 12 14 1 0 2 13: 52749 2017-08-29 13 14 0 NA NA 14: 52749 2017-10-10 14 14 0 NA NA 15: 46760 2008-01-01 1 15 1 3 6 16: 46760 2010-07-19 2 15 1 3 6 17: 46760 2011-01-14 3 15 0 NA NA 18: 46760 2011-08-02 4 15 1 3 5 19: 46760 2011-08-02 5 15 0 NA NA 20: 46760 2012-02-01 6 15 1 3 5 21: 46760 2012-02-01 7 15 0 NA NA 22: 46760 2015-04-28 8 15 1 3 7 23: 46760 2015-10-19 9 15 0 NA NA 24: 46760 2016-05-16 10 15 1 2 5 25: 46760 2016-12-22 11 15 1 1 4 26: 46760 2016-12-23 12 15 0 NA NA 27: 46760 2017-05-16 13 15 0 NA NA 28: 46760 2017-11-15 14 15 1 0 1 29: 46760 2018-02-22 15 15 0 NA NA id date idx count event num_event_5yr_fu num_subevents
中间结果是
tmp
id date date event V2 N 1: 52749 2007-01-30 2012-01-30 1 4 8 2: 52749 2007-11-27 2012-11-27 1 3 6 3: 52749 2008-10-09 2013-10-09 1 2 4 4: 52749 2011-01-06 2016-01-06 1 2 3 5: 52749 2011-07-26 2016-07-26 1 1 2 6: 52749 2015-01-12 2020-01-12 1 2 4 7: 52749 2016-09-13 2021-09-13 1 1 3 8: 52749 2017-03-21 2022-03-21 1 0 2 9: 46760 2008-01-01 2013-01-01 1 3 6 10: 46760 2010-07-19 2015-07-19 1 3 6 11: 46760 2011-08-02 2016-08-02 1 3 5 12: 46760 2012-02-01 2017-02-01 1 3 5 13: 46760 2015-04-28 2020-04-28 1 3 7 14: 46760 2016-05-16 2021-05-16 1 2 5 15: 46760 2016-12-22 2021-12-22 1 1 4 16: 46760 2017-11-15 2022-11-15 1 0 1
它仅包含event == 1的结果。在最后的 update join 中，要加入的密钥中包含event。对于具有event == 1的行，没有匹配项，因此新列将自动设置为NA。
dt = data.table(id=c(rep(52749, 14), rep(46760, 15)),
date=c("2007-01-30","2007-03-15","2007-11-27",
"2007-11-29","2008-10-09","2009-04-02",
"2011-01-06","2011-07-26","2012-01-25",
"2015-01-12","2016-09-13","2017-03-21",
"2017-08-29","2017-10-10","2008-01-01",
"2010-07-19","2011-01-14","2011-08-02",
"2011-08-02","2012-02-01","2012-02-01",
"2015-04-28","2015-10-19","2016-05-16",
"2016-12-22","2016-12-23","2017-05-16",
"2017-11-15","2018-02-22"),
idx=c(seq_len(14), seq_len(15)),
count=c(rep(14,14),rep(15,15)),
event=c(1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 0, 1,
1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 0, 0, 1, 0))
expected <-
fread("id date idx count event num_event_5yr_fu num_subevents
52749 2007-01-30 1 14 1 4 8
52749 2007-03-15 2 14 0 NA NA
52749 2007-11-27 3 14 1 3 6
52749 2007-11-29 4 14 0 NA NA
52749 2008-10-09 5 14 1 2 4
52749 2009-04-02 6 14 0 NA NA
52749 2011-01-06 7 14 1 2 3
52749 2011-07-26 8 14 1 1 2
52749 2012-01-25 9 14 0 NA NA
52749 2015-01-12 10 14 1 2 4
52749 2016-09-13 11 14 1 1 3
52749 2017-03-21 12 14 1 0 2
52749 2017-08-29 13 14 0 NA NA
52749 2017-10-10 14 14 0 NA NA
46760 2008-01-01 1 15 1 3 6
46760 2010-07-19 2 15 1 3 6
46760 2011-01-14 3 15 0 NA NA
46760 2011-08-02 4 15 1 3 6
46760 2011-08-02 5 15 0 NA NA
46760 2012-02-01 6 15 1 3 6
46760 2012-02-01 7 15 0 NA NA
46760 2015-04-28 8 15 1 3 7
46760 2015-10-19 9 15 0 NA NA
46760 2016-05-16 10 15 1 2 5
46760 2016-12-22 11 15 1 1 4
46760 2016-12-23 12 15 0 NA NA
46760 2017-05-16 13 15 0 NA NA
46760 2017-11-15 14 15 1 0 1
46760 2018-02-22 15 15 0 NA NA")[
, date := as.Date(date)]