使用dtplyr时recode函数触发‘零长度变量名’错误的求助

问题重现 #

样本数据 #

ACCOUNT <- c("M191212", "M205668", "M197954", "M208966", "M193465", "M207622", "M191869")
PRODUCT <- c(1, 2, 3, 1, 2, 3, 2)
data <- tibble(ACCOUNT, PRODUCT)

错误代码 #

library(tidyverse)
library(data.table)
library(dtplyr)

data2 <- data %>%
  lazy_dt() %>%
  mutate(YR = substring(ACCOUNT,2,3)) %>%
  mutate(PRODUCT_NAME = dplyr::recode(PRODUCT,
                         `1` = "PROD_A",
                         `2` = "PROD_B",
                         `3` = "PROD_C")) %>%
collect()

报错信息 #

Error in env_has(env, var, inherit = TRUE) : attempt to use zero-length variable name

原因分析 #

dtplyr 1.2.1版本在处理lazy_dt对象时，对dplyr::recode的整数键解析存在兼容性问题。dtplyr会将dplyr语法转换为data.table语法，而recode的整数键在转换过程中出现解析错误，生成了无效的零长度变量名。

解决方案 #

方法1：用case_when替代recode #

case_when在dtplyr中兼容性更好，逻辑直观：

data2 <- data %>%
  lazy_dt() %>%
  mutate(YR = substring(ACCOUNT,2,3)) %>%
  mutate(PRODUCT_NAME = case_when(
    PRODUCT == 1 ~ "PROD_A",
    PRODUCT == 2 ~ "PROD_B",
    PRODUCT == 3 ~ "PROD_C"
  )) %>%
collect()

方法2：使用data.table原生替换语法 #

直接利用data.table的fcase函数，适配lazy_dt的转换逻辑：

data2 <- data %>%
  lazy_dt() %>%
  mutate(YR = substring(ACCOUNT,2,3)) %>%
  .[, PRODUCT_NAME := fcase(
    PRODUCT == 1, "PROD_A",
    PRODUCT == 2, "PROD_B",
    PRODUCT == 3, "PROD_C"
  )] %>%
collect()

方法3：先转换为常规数据集再使用recode #

如果数据集规模不大，可以先将lazy_dt转为tibble，再执行recode操作：

data2 <- data %>%
  lazy_dt() %>%
  mutate(YR = substring(ACCOUNT,2,3)) %>%
  collect() %>%
  mutate(PRODUCT_NAME = dplyr::recode(PRODUCT,
                         `1` = "PROD_A",
                         `2` = "PROD_B",
                         `3` = "PROD_C"))

内容的提问来源于stack exchange，提问作者Pendragon