Question

我想创建一个dplyr::bind_rows的加强版本，当我们尝试合并的dfs中存在因子列时，它会避免Unequal factor levels: coercing to character警告（也可能包含非因子列））。这是一个例子：

df1 <- dplyr::data_frame(age = 1:3, gender = factor(c("male", "female", "female")), district = factor(c("north", "south", "west")))
df2 <- dplyr::data_frame(age = 4:6, gender = factor(c("male", "neutral", "neutral")), district = factor(c("central", "north", "east")))

然后bind_rows_with_factor_columns(df1, df2)返回（没有警告）：

dplyr::data_frame(
  age = 1:6,
  gender = factor(c("male", "female", "female", "male", "neutral", "neutral")),
  district = factor(c("north", "south", "west", "central", "north", "east"))
)

这是我到目前为止所拥有的：

bind_rows_with_factor_columns <- function(...) {
  factor_columns <- purrr::map(..., function(df) {
      colnames(dplyr::select_if(df, is.factor))
  })

  if (length(unique(factor_columns)) > 1) {
      stop("All factor columns in dfs must have the same column names")
  }

  df_list <- purrr::map(..., function (df) {
    purrr::map_if(df, is.factor, as.character) %>% dplyr::as_data_frame()
  })

  dplyr::bind_rows(df_list) %>%
    purrr::map_at(factor_columns[[1]], as.factor) %>%
    dplyr::as_data_frame()
}

我想知道是否有人对如何合并forcats包以避免必须将因素强加给角色，或者是否有人提出任何建议以提高其性能同时保持相同的功能（我想坚持tidyverse语法）。谢谢！

Answer 1

根据朋友的一个很好的解决方案来回答我自己的问题：

bind_rows_with_factor_columns <- function(...) {
  purrr::pmap_df(list(...), function(...) {
    cols_to_bind <- list(...)
    if (all(purrr::map_lgl(cols_to_bind, is.factor))) {
      forcats::fct_c(cols_to_bind)
    } else {
      unlist(cols_to_bind)
    }
  })
}

Answer 2

使用dplyr::bind_rows抑制警告，然后将所有新字符列转换回因子可能更简单。这样做的好处是可以通过列名绑定data.frames（允许列的不同排序和包含额外列），并且当因子变量有时记录为字符时仍然有效。

library(tidyverse)

bind_rows_keep_factors <- function(...) {
  ## Identify all factors
  factors <- unique(unlist(
    map(list(...), ~ select_if(..., is.factor) %>% names())
  ))
  ## Bind dataframes, convert characters back to factors
  suppressWarnings(bind_rows(...)) %>% 
    mutate_at(vars(one_of(factors)), factor)  
}

dat1 <- tibble(
  id = 1:2,
  fruit = factor(c("banana", "apple"))
)

dat2 <- tibble(
  id = 3:4,
  fruit = c("pear", "banana"),
  taste = c("Mmmm", "yum")
)

bind_rows_keep_factors(dat1, dat2)
# A tibble: 4 x 3
     id fruit  taste
  <int> <fct>  <chr>
1     1 banana NA   
2     2 apple  NA   
3     3 pear   Mmmm 
4     4 banana yum

当然，因子级别的排序被中断（恢复为字母顺序）。

使用某些因子列绑定数据帧行

2 个答案: