用 dplyr 中因子變量的模態值替換 NA 值

Question

假設我有以下data.frame。 我想用最常見的響應替換 NA， a

df <- read.table(text = "id result
1 a
2 a
3 a
4 b
5 NA", header = T)

我正在尋找這樣的東西：

calculate_mode <- function(x) {
  uniqx <- unique(x)
  uniqx[which.max(tabulate(match(x, uniqx)))]
}

df = df %>% 
  mutate(result = ifelse(is.na(result), calculate_mode(result), result))

但我不確定在定義自定義 function 之外是否有更“整潔”的方式來執行此操作。

Answer 1

  library(dplyr)
  library(tidyr)
  
  # manually get the most frequent values and tidyr::replace_na 
  most_value <- table(df$result) %>% sort(decreasing = TRUE) %>%
    head(1) %>% names()
  df %>% replace_na(list(result = most_value))
#>   id result
#> 1  1      a
#> 2  2      a
#> 3  3      a
#> 4  4      b
#> 5  5      a

動態應用於多列

  # do it acorss multiple column - still kind of using functions
  most <- function(x) {
    table(x) %>% sort(decreasing = TRUE) %>% head(1) %>% names()
  }
  multiple_column <- left_join(df, df, by = "id")
  multiple_column
#>   id result.x result.y
#> 1  1        a        a
#> 2  2        a        a
#> 3  3        a        a
#> 4  4        b        b
#> 5  5     <NA>     <NA>
  
  multiple_column %>%
    mutate(across(.cols = starts_with("result"), .fns = function(x) {
      if_else(is.na(x), most(x), x)
    }))
#>   id result.x result.y
#> 1  1        a        a
#> 2  2        a        a
#> 3  3        a        a
#> 4  4        b        b
#> 5  5        a        a

^{由代表 package (v2.0.0) 於 2021 年 4 月 24 日創建}

Answer 2

不短但可能很整潔：

library(dplyr)

df %>%
  count(result, sort = TRUE) %>%
  slice(1) %>%
  rename(mode_value = result) %>%
  select(-n) %>%
  bind_cols(df, .) %>%
  mutate(result = coalesce(result, mode_value))

#  id result mode_value
#1  1      a          a
#2  2      a          a
#3  3      a          a
#4  4      b          a
#5  5      a          a