R 自动化重编码数值型分类变量为有意义因子变量的实现方法

宽格式映射表转长格式 #

直接使用tidyverse套件的pivot_longer函数完成结构转换，处理后输出每行对应单个变量的单条数值-标签映射规则的长表：

# 加载依赖包
library(tidyverse)

# 原始宽映射表
mapping <- data.frame("Variable" = c("Var1", "Var2", "Var3", "Var4"),
                      "Value1" = c(2,2,2,7),
                      "Word1" = c("yes","yes","yes","condition A"),
                      "Value2" = c(1,1,1,6),
                      "Word2" = c("no","no","no","Condition B"),
                      "Value3" = c(8,8,8,5),
                      "Word3" = c("unk","unk","unk", "Condition C"),
                      "Value4" = c(NA,NA,NA,4),
                      "Word4" = c(NA,NA,NA,"Condition B")
)

# 宽转长核心逻辑
mapping_long <- mapping %>%
  group_by(Variable) %>%
  pivot_longer(
    cols = -Variable,
    names_to = c(".value", "seq"),
    names_pattern = "(Value|Word)(\\d+)"
  ) %>%
  drop_na(Value, Word) %>%
  ungroup() %>%
  select(-seq)

转换后的mapping_long结构示例：

批量重编码原始数据 #

拿到长格式映射表后，可直接批量完成所有分类变量的因子转换：

# 示例原始数据
data <- data.frame("ID" = c(1:5),
                   "Var1" = c(2,2,8,1,8),
                   "Var2" = c(5,8,4,88,10))

# 批量重编码核心逻辑
data_recoded <- data %>%
  mutate(across(
    # 仅处理映射表中存在的变量
    any_of(unique(mapping_long$Variable)),
    ~ factor(
      x = .,
      # 取当前变量对应的数值作为因子水平
      levels = mapping_long$Value[mapping_long$Variable == cur_column()],
      # 取当前变量对应的标签作为因子显示值
      labels = mapping_long$Word[mapping_long$Variable == cur_column()]
    )
  ))

如果需要处理映射表中未覆盖的数值（比如示例中Var2的5、88等），可提前在映射表中对应变量下补充Value=其他值, Word=unknown的规则，或在转换后用forcats::fct_explicit_na将未匹配的值统一设置为指定标签。

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