dplyr::mutate（分配 na.rm =TRUE）

Question

我有一个包含 100 个变量的 data.frame。我只想使用mutate(not summarise)获得三个变量的总和。

如果 3 个变量中的任何一个中有 NA，我仍然想获得sum. 为了做到这一点mutate，我NA用0using替换了所有值，ifelse然后我得到了sum.

library(dplyr)
df %>% mutate(mod_var1 = ifelse(is.na(var1), 0, var1),
              mod_var2 = ifelse(is.na(var2), 0, var2),
              mod_var3 = ifelse(is.na(var3), 0, var3),
              sum = (mod_var1+mod_var2+mod_var3))

有没有更好（更短）的方法来做到这一点？

数据

df <- read.table(text = c("
var1    var2    var3
4   5   NA
2   NA  3
1   2   4
NA  3   5
3   NA  2
1   1   5"), header =T)

Answer 1

rowwise()是我的首选功能。这就像group_by()但它将每一行视为一个单独的组。

df %>% rowwise() %>% mutate(Sum = sum(c(var1, var2, var3), na.rm = TRUE))

Answer 2

我们可以使用Reduce与+

df %>% 
     mutate_each(funs(replace(., is.na(.), 0)), var1:var3) %>% 
     mutate(Sum = Reduce(`+`, .))      
#   var1 var2 var3 Sum
#1    4    5    0   9
#2    2    0    3   5
#3    1    2    4   7
#4    0    3    5   8
#5    3    0    2   5
#6    1    1    5   7

或与 rowSums

df %>% 
   mutate(Sum = rowSums(.[names(.)[1:3]], na.rm = TRUE))
#   var1 var2 var3 Sum
#1    4    5   NA   9
#2    2   NA    3   5
#3    1    2    4   7
#4   NA    3    5   8
#5    3   NA    2   5
#6    1    1    5   7

基准

set.seed(24)
df1 <- as.data.frame(matrix(sample(c(NA, 1:5), 1e6 *3, replace=TRUE),
                dimnames = list(NULL, paste0("var", 1:3)), ncol=3))
system.time({
df1 %>% rowwise() %>% mutate(Sum = sum(c(var1, var2, var3), na.rm = TRUE))
})
# user  system elapsed 
#  21.50    0.03   21.66 

system.time({
df1 %>%
    mutate(rn = row_number()) %>%
    gather(var, varNum, var1:var3) %>%
    group_by(rn) %>%
    mutate(sum = sum(varNum, na.rm = TRUE)) %>% 
    spread(var, varNum)})
 # user  system elapsed 
 #  5.96    0.39    6.37 


system.time({
replace(df1, is.na(df1), 0) %>% mutate(sum = var1 + var2 + var3)
})

# user  system elapsed 
#   0.17    0.01    0.19 

system.time({
df1 %>% 
     mutate_each(funs(replace(., is.na(.), 0)), var1:var3) %>% 
     mutate(Sum = Reduce(`+`, .))      
})
# user  system elapsed 
#   0.10    0.02    0.11 

system.time({
df1 %>% 
   mutate(Sum = rowSums(.[names(.)[1:3]], na.rm = TRUE))
   })
# user  system elapsed 
#   0.04    0.00    0.03