如何在data.table中提前指定median_duration列为numeric类型？

背景与代码 #

我正在Datacamp平台学习由Arun Srinivasan和Matt Dowle讲授的《R语言data.table数据处理》课程，使用的版本为bikeshare14 0.1.4和data.table 1.14.8，编写了如下代码：

# Libraries
library(data.table)
library(bikeshare14)

# Data sets
batrips <- as.data.table(x = batrips)

# Data manipulation
batrips[,
        `:=` (median_duration = median(duration, na.rm = TRUE)),
        by = .(start_station, end_station)] |>
  _[,`:=` (trip_category = fcase(
    median_duration < 600, "short",
    median_duration >= 600 & median_duration <= 1800, "medium",
    median_duration > 1800, "long",
    default = NA))]
batrips

警告信息 #

运行后收到提示：There were 50 or more warnings (use warnings() to see the first 50)，查看警告发现均类似：

Group 11 column 'median_duration': 481.500000 (type 'double') at RHS position 1 truncated (precision lost) when assigning to type 'integer'

解决方法 #

方法1：提前初始化列并指定类型 #

在分组计算前，先为batrips添加median_duration列并设置为numeric类型，后续赋值时会保留该类型：

# 提前初始化numeric类型的列
batrips[, median_duration := as.numeric(NA)]

# 再执行分组计算赋值
batrips[,
        `:=` (median_duration = median(duration, na.rm = TRUE)),
        by = .(start_station, end_station)] |>
  _[,`:=` (trip_category = fcase(
    median_duration < 600, "short",
    median_duration >= 600 & median_duration <= 1800, "medium",
    median_duration > 1800, "long",
    default = NA))]

方法2：赋值时显式转换结果类型 #

直接在计算中位数时用as.numeric()包裹结果，确保返回值为numeric类型，避免data.table自动推断为integer：

batrips[,
        `:=` (median_duration = as.numeric(median(duration, na.rm = TRUE))),
        by = .(start_station, end_station)] |>
  _[,`:=` (trip_category = fcase(
    median_duration < 600, "short",
    median_duration >= 600 & median_duration <= 1800, "medium",
    median_duration > 1800, "long",
    default = NA))]

方法3：事后修改列类型（可选） #

如果已经完成赋值操作，可使用setcoltypes()函数修改列类型：

# 先完成原计算
batrips[,
        `:=` (median_duration = median(duration, na.rm = TRUE)),
        by = .(start_station, end_station)] |>
  _[,`:=` (trip_category = fcase(
    median_duration < 600, "short",
    median_duration >= 600 & median_duration <= 1800, "medium",
    median_duration > 1800, "long",
    default = NA))]

# 修改列类型为numeric
setcoltypes(batrips, "median_duration", "numeric")

原因说明 #

警告产生的核心原因是：原duration列是integer类型，当分组计算中位数时，部分组存在偶数个样本，中位数为小数（返回double类型），但data.table默认会尝试将新创建的median_duration列推断为integer类型，导致小数被截断，触发精度丢失警告。通过提前指定或显式转换类型，就能避免这个问题。

内容的提问来源于stack exchange，提问作者luifrancgom