为什么 dplyr 在数据聚合方面比 plyr 慢？

Question

背景问题：

假设我们有这样一个数据集：

ID DRIVE_NUM FLAG
 1         A PASS
 2         A FAIL
 3         A PASS
-----------------
 4         B PASS
 5         B PASS
 6         B PASS
-----------------
 7         C PASS
 8         C FAIL
 9         C FAIL

我想按以下规则聚合 DRIVE_NUM 的数据集：

For a specific DRIVE_NUM group,

If there is any FAIL flag in the DRIVE_NUM group, I want the first row with the FAIL flag.

If there is no FAIL flag in the group, just take the first row in the group.

所以，我将得到以下套装：

  ID DRIVE_NUM FLAG
   2         A FAIL
   4         B PASS
   8         C FAIL

更新：

看来dplyr解法比plyr还要慢。我有没有使用不当？

#Simulate Data

X = data.frame(
  group = rep(paste0("NO",1:10000),each=2),
  flag = sample(c("F","P"),20000,replace = TRUE),
  var = rnorm(20000)
)



library(plyr)
library(dplyr)

#plyr

START = proc.time()
X2 = ddply(X,.(flag),function(df) {
  if( sum(df$flag=="F")> 0){
    R = df[df$flag=="F",]
    if(nrow(R)>1) {R = R[1,]} else {R = R}
  } else{
    R = df[1,]
  }
  R
})
proc.time() - START   

#user  system elapsed 
#0.03    0.00    0.03 

#dplyr method 1

START = proc.time()
X %>%
  group_by(group) %>% 
  slice(which.min(flag))
proc.time() - START  

#user  system elapsed 
#0.22    0.02    0.23 

#dplyr method 2

START = proc.time()
X %>%
  group_by(group, flag) %>%
  slice(1) %>%
  group_by(group) %>% 
  slice(which.min(flag))
proc.time() - START  

#user  system elapsed 
#0.28    0.00    0.28

有没有 data.table 版本可以比 plyr 快得多？

Answer 1

使用data.table

library(data.table)
START = proc.time()
 X3 = as.data.table(X)[X[, .I[which.min(flag)] , by = group]$V1]
proc.time() - START
#   user  system elapsed 
#  0.00    0.02    0.02

或使用order

START = proc.time()
 X4 = as.data.table(X)[order(flag), .SD[1L] , by = group]
proc.time() - START
#    user  system elapsed 
#    0.02    0.00    0.01

使用OP代码的dplyr和plyr对应的时间是

#   user  system elapsed 
#  0.28    0.04    2.68 

#   user  system elapsed 
#  0.01    0.06    0.67

另外正如@Frank 所评论的，base R 方法时间是

START = proc.time()
Z = X[order(X$flag),]
X5 = with(Z, Z[tapply(seq(nrow(X)), group, head, 1), ])
proc.time() - START
#    user  system elapsed 
#    0.15    0.03    0.65

我猜 slice 正在减慢 dplyr。

Answer 2

好吧，这并不比 data.table 快，但绝对是一个进步：

START = proc.time()
m3 <- X %>%
    group_by(group) %>% 
    arrange(flag) %>%
    slice(1)
proc.time() - START

#user  system elapsed 
#0.03    0.00    0.05 

# OP - method 1
START = proc.time()
m1 <- X %>%
    group_by(group) %>% 
    slice(which.min(flag))
proc.time() - START

#user  system elapsed 
#0.31    0.00    0.33 

# OP - method 2
START = proc.time()
m2 <- X %>%
    group_by(group, flag) %>%
    slice(1) %>%
    group_by(group) %>% 
    slice(which.min(flag))
proc.time() - START 

#user  system elapsed 
#0.39    0.02    0.45 

identical(m2, m3)
[1] TRUE

为什么 dplyr 在数据聚合方面比 plyr 慢？

Why is dplyr slower than plyr for data aggregation?

r

aggregation

plyr

dplyr