为何dpplyr分组函数添加max后出现异常结果？

问题背景 #

运行以下dplyr代码时，GroupRank列能得到预期分组结果；但将GroupRank = if_else(as.character(Group) == "0", ElementCnt, min(GroupRank))修改为GroupRank = max(1L,if_else( as.character(Group) == "0", ElementCnt, min(GroupRank)))后，Group为0的行GroupRank均变为3，与预期不符。

初始代码及预期结果 #

library(dplyr)

myData <- 
  data.frame(
    Element = c("A","A","B","A","C","C"),
    Group = c(0,0,0,0,1,1)
  )

myDataGroups <- myData %>%
  mutate(origOrder = row_number()) %>%  
  group_by(Element) %>% 
  mutate(ElementCnt = row_number()) %>%
  ungroup() %>%  
  mutate(Group = factor(Group, unique(Group))) %>% 
  arrange(Group) %>% 
  mutate(groupCt = cumsum(Group != lag(Group, 1, Group[[1]])) - 1L) %>%  
  group_by(Group) %>%  
  mutate(GroupRank = ElementCnt - max(0L,groupCt),
         GroupRank = if_else(as.character(Group) == "0", ElementCnt, min(GroupRank))
  )%>%  
  ungroup() %>%
  arrange(origOrder)
myDataGroups

预期输出：

# A tibble: 6 x 6
  Element Group origOrder ElementCnt groupCt GroupRank
  <chr>   <fct>     <int>      <int>   <int>     <int>
1 A       0             1          1      -1         1
2 A       0             2          2      -1         2
3 B       0             3          1      -1         1
4 A       0             4          3      -1         3
5 C       1             5          1       0         1
6 C       1             6          2       0         1

修改后代码的异常结果 #

修改后的关键行：

GroupRank = max(1L,if_else( as.character(Group) == "0", ElementCnt, min(GroupRank)))

异常输出：

Element Group origOrder ElementCnt groupCt GroupRank
  <chr>   <fct>     <int>      <int>   <int>     <int>
1 A       0             1          1      -1         3
2 A       0             2          2      -1         3
3 B       0             3          1      -1         3
4 A       0             4          3      -1         3
5 C       1             5          1       0         1
6 C       1             6          2       0         1

错误原因 #

问题出在max()函数的使用方式：

• 在dplyr的group_by上下文里，max()是对整个分组的向量计算全局最大值，而非逐行进行元素级比较。
• 对于Group=0的分组，if_else(as.character(Group) == "0", ElementCnt, min(GroupRank))返回的是该组的ElementCnt向量c(1,2,1,3)，max(1L, 这个向量)会直接取整个向量的最大值3，然后将这个值广播到组内所有行，导致所有Group=0的行GroupRank都变成3。

解决方法 #

要实现逐行取最大值的需求，需要使用元素级别的最大值函数pmax()，它会对输入的向量逐元素比较并返回对应位置的最大值，而非全局最大。

修改后的关键代码行：

GroupRank = pmax(1L, if_else(as.character(Group) == "0", ElementCnt, min(GroupRank)))

修正后的完整代码及输出 #

library(dplyr)

myData <- 
  data.frame(
    Element = c("A","A","B","A","C","C"),
    Group = c(0,0,0,0,1,1)
  )

myDataGroups <- myData %>%
  mutate(origOrder = row_number()) %>%  
  group_by(Element) %>% 
  mutate(ElementCnt = row_number()) %>%
  ungroup() %>%  
  mutate(Group = factor(Group, unique(Group))) %>% 
  arrange(Group) %>% 
  mutate(groupCt = cumsum(Group != lag(Group, 1, Group[[1]])) - 1L) %>%  
  group_by(Group) %>%  
  mutate(GroupRank = ElementCnt - max(0L,groupCt),
         GroupRank = pmax(1L, if_else(as.character(Group) == "0", ElementCnt, min(GroupRank)))
  )%>%  
  ungroup() %>%
  arrange(origOrder)
myDataGroups

输出结果（符合预期）：

# A tibble: 6 x 6
  Element Group origOrder ElementCnt groupCt GroupRank
  <chr>   <fct>     <int>      <int>   <int>     <int>
1 A       0             1          1      -1         1
2 A       0             2          2      -1         2
3 B       0             3          1      -1         1
4 A       0             4          3      -1         3
5 C       1             5          1       0         1
6 C       1             6          2       0         1

内容的提问来源于stack exchange，提问作者Curious Jorge - user9788072