dplyr,data.table和setDT交互问题

Question

这是一个更大规模的问题的简化版本.目标是使用data.table结构和dplyr命令更快地对多列进行排序和分组.

正确的版本如下:

library(dplyr)
library(data.table)
library(dtplyr)
library(lubridate)

# data set
dt = data.frame(id = c("a","b", "a"),
                date = ymd(c("2016-01-03","2016-01-02","2016-01-01")),
                value = c(10,5,9), stringsAsFactors = F)

# process to get the id of the largest value
(setDT(dt, key=c("id","value")) %>% select(id,value) %>% arrange(desc(value)) %>% slice(1))$id -> picked_id

# return all rows of this id
dt %>% filter(id %in% picked_id)

# id          date value
# 1:  a 2016-01-01     9
# 2:  a 2016-01-03    10

但是当我尝试setDT在我的脚本中使用不同的位置时,我得到了不同的结果:

dt = data.frame(id = c("a","b", "a"),
                date = ymd(c("2016-01-03","2016-01-02","2016-01-01")),
                value = c(10,5,9), stringsAsFactors = F)

(dt %>% select(id,value) %>% setDT(., key=c("id","value")) %>% arrange(desc(value)) %>% slice(1))$id -> picked_id

dt %>% filter(id %in% picked_id)

#   id       date value
# 1  a 2016-01-03     9
# 2  a 2016-01-02    10

显然,对于这个简单的任务,还有其他更容易理解的脚本,但我想了解为什么会出现此问题.

Answer 1

您不能安全地混合(i)data.table的函数,这些函数通过引用修改为(ii)dplyr链,其设计为永远不会通过引用进行修改.看看这里发生了什么:

library(dplyr)
library(data.table)
library(dtplyr)
library(lubridate)

dt = data.frame(id = c("a","b", "a"),
                date = ymd(c("2016-01-03","2016-01-02","2016-01-01")),
                value = c(10,5,9), stringsAsFactors = FALSE)

dt


  id       date value
1  a 2016-01-03    10
2  b 2016-01-02     5
3  a 2016-01-01     9


dt %>% select(id,value) %>% setDT(., key=c("id","value"))

dt


  id       date value
1  a 2016-01-03     9
2  a 2016-01-02    10
3  b 2016-01-01     5

因此selected列已被setDT调用修改.您可以将此视为dtplyr执行中的错误select或OP的滥用.无论如何,我会一次坚持使用一个范例(就个人而言,我只是使用带有magrittr的data.table而且从未遇到过这些问题).现在,您可以添加一个copy链,

dt %>% select(id,value) %>% copy %>% setDT(., key=c("id","value"))

但我想你需要在整个地方做到这一点.