R中的滞后函数。计算数据框中先前发生的事件

Question

我是 R 新手。 我有一个这样的数据框：

p_id    start_date  ch  end_date
5713729 01/10/2014  1   20/03/2015
5713729 01/04/2016  0   NA
5713731 01/12/2010  1   03/02/2012
5713731 01/04/2013  1   30/10/2014
5713731 01/01/2015  0   NA
5713735 01/07/2012  0   NA
5713736 01/07/2007  1   30/06/2012
5713736 01/04/2016  0   NA
5713737 01/06/2016  0   NA

我需要计算每个 p_id，每行中有多少次事件“ch”。 所以数据框必须按 p_id 和日期（asc）排序。 首先我尝试了 whit ifelse 函数：

#sort
library(dplyr)
data <- data %>% arrange(p_id,start_date,end_date)

#initialize count:
data$count_ch_prev <- 0

#count (not good...)
data$count_ch_prev <- ifelse(data$p_id == 
lag(data$p_id,1),lag(data$count_ch_prev,1) + 
lag(data$ch,1),data$count_ch_prev)

结果是：

p_id    start_date  ch  end_date    count_ch_prev
5713729 01/10/2014  1   20/03/2015  NA
5713729 01/04/2016  0   NA          1
5713731 01/12/2010  1   03/02/2012  0
5713731 01/04/2013  1   30/10/2014  1
5713731 01/01/2015  0   NA          1
5713735 01/07/2012  0   NA          0
5713736 01/07/2007  1   30/06/2012  0
5713736 01/04/2016  0   NA          1
5713737 01/06/2016  0   NA          0    

寻找类似的问题 (Lag doesn't see the effects of mutate on previous rows)，我意识到这个函数是矢量化的，所以它不会逐行计算。相反，它同时计算所有行。

我的预期结果是这样的：

p_id    start_date  ch  end_date    count_ch_prev
5713729 01/10/2014  1   20/03/2015  0
5713729 01/04/2016  0   NA          1
5713731 01/12/2010  1   03/02/2012  0
5713731 01/04/2013  1   30/10/2014  1
5713731 01/01/2015  0   NA          2
5713735 01/07/2012  0   NA          0
5713736 01/07/2007  1   30/06/2012  0
5713736 01/04/2016  0   NA          1
5713737 01/06/2016  0   NA          0

我也尝试过使用 while 循环：

data$count_ch_prev <- 0
while (data$p_id == lag(data$p_id,1)) {
data$count_ch_prev <- lag(data$count_ch_prev) + lag(data$ch)
}

但我得到了相同的“整体”结果。我必须使用哪个功能？

要复制的代码：

p_id <- 
c(5713729,5713729,5713731,5713731,5713731,5713735,5713736,5713736,5713737)
start_date <- as.Date(c('2014-10-01','2016-04-01','2010-12-01','2013-04- 
01','2015-01-01','2012-07-01','2007-07-01','2016-04-01','2016-06-01'))
end_date <- as.Date(c('2015-03-20',NA,'2012-02-03','2014-10-30',NA,NA,'2012- 
06-30',NA,NA))
ch <- c(1,0,1,1,0,0,1,0,0)
data <- data.frame(p_id,start_date,ch,end_date)

Answer 1

我认为你可以使用dplyr 来分组p_id，然后使用lag 和cumsum：

library(dplyr)

data %>%
  group_by(p_id) %>%
  mutate(count_ch_prev = lag(cumsum(ch), default = 0))

输出：

# A tibble: 9 x 5
# Groups:   p_id [5]
     p_id start_date    ch end_date   count_ch_prev
    <dbl> <date>     <dbl> <date>             <dbl>
1 5713729 2014-10-01     1 2015-03-20             0
2 5713729 2016-04-01     0 NA                     1
3 5713731 2010-12-01     1 2012-02-03             0
4 5713731 NA             1 2014-10-30             1
5 5713731 2015-01-01     0 NA                     2
6 5713735 2012-07-01     0 NA                     0
7 5713736 2007-07-01     1 NA                     0
8 5713736 2016-04-01     0 NA                     1
9 5713737 2016-06-01     0 NA                     0

数据表替代：

library(data.table)

dt <- data.table(data)

dt[, count_ch_prev := shift(cumsum(ch), fill = 0), by = p_id]

输出：

> dt
      p_id start_date ch   end_date count_ch_prev
1: 5713729 2014-10-01  1 2015-03-20             0
2: 5713729 2016-04-01  0       <NA>             1
3: 5713731 2010-12-01  1 2012-02-03             0
4: 5713731       <NA>  1 2014-10-30             1
5: 5713731 2015-01-01  0       <NA>             2
6: 5713735 2012-07-01  0       <NA>             0
7: 5713736 2007-07-01  1       <NA>             0
8: 5713736 2016-04-01  0       <NA>             1
9: 5713737 2016-06-01  0       <NA>             0