Question

我有一个包含250,000行的data.table。一小部分样本如下所示：

library('data.table')
stack <- structure(list(ID = c(1, 2, 3, 4, 
                              5, 6), 
                        Special = c(FALSE, FALSE, TRUE, FALSE, 
                                                                 FALSE, FALSE), 
                        ENTER = structure(c(1453797143, 1453850662, 1453444763, 1453656563, 1453410022, 1453367723), 
                                          class = c("POSIXct", "POSIXt"), tzone = ""), 
                        LEAVE = structure(c(1453803923, 1453856002, 1453450403,1453657823, 1453418123, 1453377382), 
                                          class = c("POSIXct", "POSIXt"), tzone = "")), 
                   .Names = c("ID", "Special", "ENTER", "LEAVE"), 
                   class = c("data.table", "data.frame"),
                   row.names = c(NA, -6L))

stack
   ID Special               ENTER               LEAVE
1:  1   FALSE 2016-01-26 09:32:23 2016-01-26 11:25:23
2:  2   FALSE 2016-01-27 00:24:22 2016-01-27 01:53:22
3:  3    TRUE 2016-01-22 07:39:23 2016-01-22 09:13:23
4:  4   FALSE 2016-01-24 18:29:23 2016-01-24 18:50:23
5:  5   FALSE 2016-01-21 22:00:22 2016-01-22 00:15:23
6:  6   FALSE 2016-01-21 10:15:23 2016-01-21 12:56:22

我正在以每小时的间隔寻找WIP（正在进行的工作）。这是已输入但尚未在特定时间离开的实例数的计数。我现在使用以下功能执行此操作：

get_WIPlevels <- function(data) {
  # determine range
  min <- round(min(data$ENTER), "hours")
  max <- round(max(data$LEAVE), "hours")
  range <- seq(from=min, to=max, by='hours')

  # create target data.table
  WIPLevels <- data.table(Timestamp = range, WIP = integer(length(range)))

  # calculate WIP values
  WIPLevels$WIP <- sapply(WIPLevels$Timestamp, function(x) 
    nrow(data[data$ENTER<=x & data$LEAVE > x,]))
  return(WIPLevels)
}

这可以按预期工作，结果如下：

get_WIPLevels(stack)
               Timestamp WIP
  1: 2016-01-21 10:00:00   0
  2: 2016-01-21 11:00:00   1
  3: 2016-01-21 12:00:00   1
  4: 2016-01-21 13:00:00   0
  5: 2016-01-21 14:00:00   0
 ---                        
137: 2016-01-27 02:00:00   0

但是，它不是最快的功能。将它应用于完整数据集只需要一分钟：

system.time(get_WIPlevels(fulldata))
   user  system elapsed 
  53.72    2.15   56.19

有什么建议可以加快速度吗？

Answer 1

您可以使用滚动连接尝试此操作

tr <- data.table(
  Timestamp = seq(lubridate::floor_date(min(stack$ENTER), "hour"), 
                  lubridate::ceiling_date(max(stack$LEAVE), "hour"), by = "hours"))
setkey(tr, Timestamp)

stack[, Timestamp := ENTER]
setkey(stack, Timestamp)

ms <- stack[tr, roll = +Inf]
ms[, .(WIP = sum(!is.na(ENTER) & LEAVE >= Timestamp)), by = Timestamp]

从时间戳计算正在进行的工作

1 个答案: