R：计算时间间隔内的日期

Question

假设我们有数据输入：

df.in <- data.frame(event = c(1,2,3,4,5), 
                    start = c("2015-01-01", "2015-01-01", "2015-01-02",
                              "2015-01-02", "2015-01-03"),
                    end = c("2015-01-03", "2015-01-04", "2015-01-03",
                            "2015-01-05", "2015-01-05"))
df.in$start <- as.Date(df.in$start, "%Y-%m-%d")
df.in$end <- as.Date(df.in$end, "%Y-%m-%d")

> df.in
  event      start        end
1     1 2015-01-01 2015-01-03
2     2 2015-01-01 2015-01-04
3     3 2015-01-02 2015-01-03
4     4 2015-01-02 2015-01-05
5     5 2015-01-03 2015-01-05

目标是计算所有事件的发生日期（包括开始，不包括结束）。填写此数据框：

df.out <- data.frame(date = c("2015-01-01", "2015-01-02", "2015-01-03", 
                              "2015-01-04", "2015-01-05"),
                     count = 0)
df.out$date <- as.Date(df.out$date, "%Y-%m-%d")
> df.out
        date count
1 2015-01-01     0
2 2015-01-02     0
3 2015-01-03     0
4 2015-01-04     0
5 2015-01-05     0

从概念上讲，它看起来像这样：

#1 **
#2 ****
#3 ***
#4 **
#5 

所以，我目前的想法是循环：

for(i in seq_along(df.out$date)){
  temp.df <- df.in[df.in$start <= df.out$date[i],]
  df.out$count[i] <- nrow(temp.df) - nrow(temp.df[temp.df$end <= df.out$date[i],])
}
> df.out
        date count
1 2015-01-01     2
2 2015-01-02     4
3 2015-01-03     3
4 2015-01-04     2
5 2015-01-05     0

它有效，但我有点担心我正在调用的这个temp.df 可能会滚雪球成非常大的东西。鉴于事件的数量很容易达到数万甚至数十万。

所以我的问题是 - 有没有更有效的方法？也许通过使用一些日期包，例如lubridate，我可以以某种方式对整个事情进行矢量化？

Answer 1

所以我对data.table::foverlaps() 进行了研究。我会把我的发现留给任何可能觉得它有用的人，因为老实说，我在搜索类似帖子时并没有真正找到这些小东西。

鉴于我们正在比较区间，并且我们仅在 y 参数上有区间，在这种特殊情况下为 df.in - 我们必须人为地制作一个。例如df.out$date2 <- df.out$date。此外，没有直接（或我找不到任何）方法来设置包含或排除设置间隔端点。鉴于我们要排除 df.in$end 中的端点，我们必须使用简单的 df.in$end <- df.in$end - 1 在数据表本身上手动执行此操作。

长话短说，这是一个工作示例：

require(data.table)
df.out <- data.table(date = c("2015-01-01", "2015-01-02", "2015-01-03", 
                              "2015-01-04", "2015-01-05"),
                     count = 0)
df.out$date <- as.Date(df.out$date, "%Y-%m-%d")

df.in <- data.table(event = c(1,2,3,4,5), 
                    start = c("2015-01-01", "2015-01-01", "2015-01-02",
                              "2015-01-02", "2015-01-03"),
                    end = c("2015-01-03", "2015-01-04", "2015-01-03",
                            "2015-01-05", "2015-01-05"))
df.in$start <- as.Date(df.in$start, "%Y-%m-%d")
df.in$end <- as.Date(df.in$end, "%Y-%m-%d") - 1

setkey(df.in, start, end)
df.out$date2 <- df.out$date
df.test <- foverlaps(x = df.out, y = df.in, type = "within", by.x = c("date", "date2"), by.y = c("start", "end"))
df.test$count[!is.na(df.test$event)] <- 1
aggregate(count ~ date, data = df.test, sum)

        date count
1 2015-01-01     2
2 2015-01-02     4
3 2015-01-03     3
4 2015-01-04     2
5 2015-01-05     0

---

或者，你可以这样做

数据

df.out <- data.table(date = as.Date(c("2015-01-01", "2015-01-02", "2015-01-03", 
                              "2015-01-04", "2015-01-05")))

df.in <- data.table(event = 1:5, 
                    start = as.Date(c("2015-01-01", "2015-01-01", "2015-01-02",
                              "2015-01-02", "2015-01-03")),
                    end = as.Date(c("2015-01-03", "2015-01-04", "2015-01-03",
                            "2015-01-05", "2015-01-05")))

解决方案

df.out[, `:=`(start = date, end = date)]
df.in[, end := end - 1L]
setkey(df.out, start, end)
foverlaps(df.in, df.out)[, .(count = .N), by = date]
#          date count
# 1: 2015-01-01     2
# 2: 2015-01-02     4
# 3: 2015-01-03     3
# 4: 2015-01-04     2

或者，如果你想更新df.out，你也可以这样做

res <- foverlaps(df.in, df.out, which = TRUE)[, .N, by = yid]
df.out[res$yid, Count := res$N]
df.out[is.na(Count), Count := 0L]