在R的data.table中标记连续0值的分组并编号

现有如下data.table数据集：

set.seed(0) 
data <- data.table(id = as.factor(c(rep("C001", 3), "C002", rep("C003", 5), rep("C004", 2), rep("C005", 7))),
                   period = as.factor(c(1, 2, 3, 2, 1, 4, 5, 6, 10, 3, 4, 2, 3, 4, 7, 8, 9, 10)),
                   industry = as.factor(c(rep("Finance", 3), "Culture", rep("Nutrition", 5), rep("Finance", 2), rep("Medicine", 7))),
                   present = rbinom(18, 1, prob = 0.3))

需要新增一列present_count，对连续的present=0分组进行编号标记：

• 当present=1时，标记为"1"
• 当present=0时，按连续组依次标记为"0_1"、"0_2"、"0_3"...

预期输出示例：

set.seed(0)
data.table(id = as.factor(c(rep("C001", 3), "C002", rep("C003", 5), rep("C004", 2), rep("C005", 7))),
           period = as.factor(c(1, 2, 3, 2, 1, 4, 5, 6, 10, 3, 4, 2, 3, 4, 7, 8, 9, 10)),
           industry = as.factor(c(rep("Finance", 3), "Culture", rep("Nutrition", 5), rep("Finance", 2), rep("Medicine", 7))),
           present = rbinom(18, 1, prob = 0.3),
           present_count = c("1", "0_1", "0_1", "0_1", "1", "0_2", "1", "1", "0_3","0_3","0_3","0_3","0_3","0_3","0_3","1","0_4", "1"))

已知用while循环可实现，但数据集规模大时效率低下，需高效实现方法。

利用data.table内置的rleid()函数（快速生成连续分组ID），结合分组赋值实现矢量化操作，避免循环，大幅提升效率：

set.seed(0) 
data <- data.table(id = as.factor(c(rep("C001", 3), "C002", rep("C003", 5), rep("C004", 2), rep("C005", 7))),
                   period = as.factor(c(1, 2, 3, 2, 1, 4, 5, 6, 10, 3, 4, 2, 3, 4, 7, 8, 9, 10)),
                   industry = as.factor(c(rep("Finance", 3), "Culture", rep("Nutrition", 5), rep("Finance", 2), rep("Medicine", 7))),
                   present = rbinom(18, 1, prob = 0.3))

# 生成连续分组ID，以present的变化为分界
data[, rleid := rleid(present)]

# 为present=0的组分配全局递增编号
data[present == 0, group_num := .GRP, by = rleid]

# 生成目标列present_count
data[, present_count := ifelse(present == 1, "1", paste0("0_", group_num))]

# 可选：删除中间辅助列
data[, c("rleid", "group_num") := NULL]

核心逻辑说明： #

• rleid(present)会为每一段连续相同的present值生成唯一ID，确保同一连续0组被分到同一ID下
• 对present=0的组按rleid分组，.GRP会自动生成全局递增的组编号
• 全程为矢量化操作，无需遍历行，处理百万级以上数据时效率远高于循环方法

内容的提问来源于stack exchange，提问作者reinoud