[R Lang] R Lang與高級醫學統計學（10）

這篇文章講因子資料的處理。關於因子資料（Factor），我已經在本系列文章《[R Lang] R Lang與高級醫學統計學（2）》中詳細闡述。

對於因子資料的處理，可以使用tidyverse套件系統中的forcats套件進行處理。

基本函式

forcats套件的基本函式包含：

• fct_count(f, sort = FALSE, prop = FALSE)：計算類別水準數目。
• fct_unique(f)：呈現專一類別水準名稱。
• fct_c(f1, f2)：合併不同類別水準的2個因子物件。

library(dplyr)
library(ggplot2)
library(forcats)
library(kableExtra)
## Warning: package 'kableExtra' was built under R version 4.0.3
set.seed(100)
letters[1:5]
## [1] "a" "b" "c" "d" "e"
f <- factor(sample(letters[1:5])[rpois(100, 5)])
table(f)
## f
##  a  b  c  d  e 
## 21  2  3 21 15
fct_count(f)
## # A tibble: 6 x 2
##   f         n
##   <fct> <int>
## 1 a        21
## 2 b         2
## 3 c         3
## 4 d        21
## 5 e        15
## 6 <NA>     38
fct_count(f, sort = TRUE)
## # A tibble: 6 x 2
##   f         n
##   <fct> <int>
## 1 <NA>     38
## 2 a        21
## 3 d        21
## 4 e        15
## 5 c         3
## 6 b         2
fct_count(f, sort = TRUE, prop = TRUE)
## # A tibble: 6 x 3
##   f         n     p
##   <fct> <int> <dbl>
## 1 <NA>     38  0.38
## 2 a        21  0.21
## 3 d        21  0.21
## 4 e        15  0.15
## 5 c         3  0.03
## 6 b         2  0.02
##
f1 <- factor(letters[1:3])
f2 <- factor(letters[c(1, 2, 23)])
f1
## [1] a b c
## Levels: a b c
f2
## [1] a b w
## Levels: a b w
fct_c(f1, f2)
## [1] a b c a b w
## Levels: a b c w

移除或增加類別水準

函式fct_drop()可以移除部分類別水準，函式fct_expand()可以增加因子變數中的類別水準。函式fct_explicit_na可以明確設性缺失值為1項類別水準。

f <- factor(c("F", "M"), levels = c("F", "M", "Other"))
f
## [1] F M
## Levels: F M Other
fct_drop(f)
## [1] F M
## Levels: F M
# Set only to restrict which levels to drop
fct_drop(f, only = "F")
## [1] F M
## Levels: F M Other
fct_drop(f, only = "Other")
## [1] F M
## Levels: F M
## 
fct_expand(f, "B", "T")
## [1] F M
## Levels: F M Other B T
##
f <- factor(c("F", "M", "M", "F", "F", "B", "T", NA, NA))
f
## [1] F    M    M    F    F    B    T    <NA> <NA>
## Levels: B F M T
fct_explicit_na(f)
## [1] F         M         M         F         F         B         T         (Missing)
## [9] (Missing)
## Levels: B F M T (Missing)
fct_explicit_na(f, na_level = "Other")
## [1] F     M     M     F     F     B     T     Other Other
## Levels: B F M T Other

改變或合併類別水準函式

系列函式fct_lump()可將部分類別水準合併。這個系列包含：

• fct_lump_min()：合併類別水準頻率計數低於設定的最小值。
• fct_lump_prop()：合併類別水準相對頻率低於設定的最小值。
• fct_lump_n()：合併類別水準最多n種主要類別。
• fct_lump_lowfreq()：合併類別水準，且確保other類別的頻率仍是最低。

函式定義如下：

fct_lump(f, n, prop, w = NULL, other_level = "Other",
         ties.method = c("min", "average", "first", "last", "random", "max"))
fct_lump_min(f, min, w = NULL, other_level = "Other")
fct_lump_prop(f, prop, w = NULL, other_level = "Other")
fct_lump_n(f, n, w = NULL, other_level = "Other",
  ties.method = c("min", "average", "first", "last", "random", "max"))
fct_lump_lowfreq(f, other_level = "Other")

其中引數f為因子向量，n設定最多n種主要類別，prop設定正值百分率，合併小於prop的類別，設定負值百分率，合併大於prop的類別。w設定權重。other_level設定合併後的類別名稱。ties.method處理相同排序方式。min保留至少出現min次類別。

x <- factor(rep(LETTERS[1:9], times = c(40, 10, 5, 27, 1, 1, 1, 1, 1)))
x %>% table()
## .
##  A  B  C  D  E  F  G  H  I 
## 40 10  5 27  1  1  1  1  1
x %>% fct_lump_n(3) %>% table()
## .
##     A     B     D Other 
##    40    10    27    10
x %>% fct_lump_prop(0.10) %>% table()
## .
##     A     B     D Other 
##    40    10    27    10
x %>% fct_lump_min(5) %>% table()
## .
##     A     B     C     D Other 
##    40    10     5    27     5
x %>% fct_lump_lowfreq() %>% table()
## .
##     A     D Other 
##    40    27    20
##
set.seed(123)
x <- factor(letters[rpois(50, 5)])
x
##  [1] d g d h i b e h e e i e f e b h c b d i h f f k f f e e d c i h f g a e f c d c c d
## [43] d d c c c e d g
## Levels: a b c d e f g h i k
table(x)
## x
## a b c d e f g h i k 
## 1 3 8 9 9 7 3 5 4 1
table(fct_lump_lowfreq(x))
## 
##     b     c     d     e     f     g     h     i Other 
##     3     8     9     9     7     3     5     4     2
## Use positive values to collapse the rarest
fct_lump_n(x, n = 3)
##  [1] d     Other d     Other Other Other e     Other e     e     Other e     Other e    
## [15] Other Other c     Other d     Other Other Other Other Other Other Other e     e    
## [29] d     c     Other Other Other Other Other e     Other c     d     c     c     d    
## [43] d     d     c     c     c     e     d     Other
## Levels: c d e Other
fct_lump_prop(x, prop = 0.1)
##  [1] d     Other d     Other Other Other e     Other e     e     Other e     f     e    
## [15] Other Other c     Other d     Other Other f     f     Other f     f     e     e    
## [29] d     c     Other Other f     Other Other e     f     c     d     c     c     d    
## [43] d     d     c     c     c     e     d     Other
## Levels: c d e f Other
## Use negative values to collapse the most common
fct_lump_n(x, n = -3)
##  [1] Other g     Other Other Other b     Other Other Other Other Other Other Other Other
## [15] b     Other Other b     Other Other Other Other Other k     Other Other Other Other
## [29] Other Other Other Other Other g     a     Other Other Other Other Other Other Other
## [43] Other Other Other Other Other Other Other g    
## Levels: a b g k Other
fct_lump_prop(x, prop = -0.1)
##  [1] Other g     Other h     i     b     Other h     Other Other i     Other Other Other
## [15] b     h     Other b     Other i     h     Other Other k     Other Other Other Other
## [29] Other Other i     h     Other g     a     Other Other Other Other Other Other Other
## [43] Other Other Other Other Other Other Other g    
## Levels: a b g h i k Other
## Use weighted frequencies
w <- c(rep(2, 25), rep(1, 25))
fct_lump_n(x, n = 5, w = w)
##  [1] d     Other d     h     Other Other e     h     e     e     Other e     f     e    
## [15] Other h     c     Other d     Other h     f     f     Other f     f     e     e    
## [29] d     c     Other h     f     Other Other e     f     c     d     c     c     d    
## [43] d     d     c     c     c     e     d     Other
## Levels: c d e f h Other
fct_lump_n(x, n = 6)
##  [1] d     Other d     h     i     Other e     h     e     e     i     e     f     e    
## [15] Other h     c     Other d     i     h     f     f     Other f     f     e     e    
## [29] d     c     i     h     f     Other Other e     f     c     d     c     c     d    
## [43] d     d     c     c     c     e     d     Other
## Levels: c d e f h i Other
fct_lump_n(x, n = 6, ties.method = "max")
##  [1] d     Other d     h     i     Other e     h     e     e     i     e     f     e    
## [15] Other h     c     Other d     i     h     f     f     Other f     f     e     e    
## [29] d     c     i     h     f     Other Other e     f     c     d     c     c     d    
## [43] d     d     c     c     c     e     d     Other
## Levels: c d e f h i Other
## Use fct_lump_min() to lump together all levels with fewer than `n` values
table(fct_lump_min(x, min = 10))
## 
## Other 
##    50

類別重新排列函式

因子資料的類別重新排序函式有：

• fct_rev(f)：將反轉原有類別出現的排列順序。
• fct_shuffle(f, n = 1L)：將原有類別出現的排列順序隨機變更。
• fct_reorder(.f, .x, .fun = median, …, .desc = FALSE)
• fct_reorder2(.f, .x, .y, .fun = last2, …, .desc = TRUE)
• first2(.x, .y)
• last2(.x, .y)

fct_reorder()將因子f類別出現的排列順序依照其他變數更動，fct_reorder2()保留因子f原有類別出現的排列順序。當y變數依照x變數排序，函式first2(.x, .y)與last2(.x, .y)可尋找y變數的最前與最後的2個數值。

引數解釋：

• .f：為主要因子變數。
• .x, .y：為其他變數。
• .fun：為摘要函式。
• .desc = FALSE：由小到大排序。

f <- factor(c("a", "b", "c"))
fct_rev(f)
## [1] a b c
## Levels: c b a
fct_shuffle(f)
## [1] a b c
## Levels: a c b
fct_shuffle(f)
## [1] a b c
## Levels: c b a
##
df <- tibble::tribble(
  ~color,     ~a, ~b,
  "blue",      1,  2,
  "green",     6,  2,
  "purple",    3,  3,
  "red",       2,  3,
  "yellow",    5,  1
)
df$color <- factor(df$color)
##
fct_reorder(df$color, df$a, min)
## [1] blue   green  purple red    yellow
## Levels: blue red purple yellow green
fct_reorder2(df$color, df$a, df$b)
## [1] blue   green  purple red    yellow
## Levels: purple red blue green yellow