R 使用data.table按分组变量查找较大或较小的值

我的源数据有几个月对应的数据，但在这些数据中，我只想比较来自预先指定月份的数据

这是我的输入数据：

dput(mydf)
structure(list(Month = structure(c(1L, 2L, 1L, 2L, 3L, 1L, 2L, 
2L, 1L, 2L, 1L), .Label = c("Aug", "Oct", "Sep"), class = "factor"), 
    Pipe = c(3, 4, 5, 3, 2, 1, 3, 3, 4, NA, 5), Gp = structure(c(1L, 
    1L, 2L, 2L, 2L, 3L, 4L, 5L, 5L, 6L, 6L), .Label = c("A", 
    "B", "C", "D", "E", "F"), class = "factor")), .Names = c("Month", 
"Pipe", "Gp"), row.names = c(NA, -11L), class = "data.frame")

现在，在这三个月中，我只想比较以下变量指定的月份

 This_month_to_compare <- "Oct"
  Last_Month_to_compare <- "Aug"

我已经在上面手动添加了解释

我确实尝试过编码，以下是我的尝试：

mydfi<-data.table::as.data.table(mydfi)
  mydf<-mydfi
  #Method 1: Convert to Wide Format
  #Convert to wide format
  mydf<-data.table::dcast(mydf,Gp ~ Month, value.var = "Pipe")
  #Compare
  mydf$Growth<-mydf[[This_month_to_compare]]>mydf[[Last_Month_to_compare]]
  #Back to long format
  Melt_columns<-c("Aug","Oct","Sep")
  mydf<-data.table::melt(mydf, measure.vars =Melt_columns,variable.name = "Month", value.name = "Pipe")
  mydfo<-mydf[mydfi,on=c("Month","Gp","Pipe")]
  mydfo[Month!=This_month_to_compare,"Growth"]<-NA

更新：我可以通过添加一个左连接来解决上述问题。我已经更新了上面的代码。但是，我正在寻找以下方面的解决方案：

原因是我的实际数据集非常大，不允许连接

---

任何帮助都将不胜感激。提前谢谢。

这就是你想的吗

> library(data.table)
> mydf <- data.table(mydf)
> This_month_to_compare <- "Oct"
> Last_Month_to_compare <- "Aug"
> setkey(mydf, Gp, Month)
> 
> # Make dummy table to join with
> mydf[
+   , Pipe_this := .SD[Month == This_month_to_compare, Pipe], by = "Gp"][
+     , Pipe_last := .SD[Month == Last_Month_to_compare, Pipe], by = "Gp"][
+       , `:=`(
+         Greater = Pipe_last < Pipe_this, Pipe_last = NULL, Pipe_this = NULL)][
+           Month != "Oct", Greater := NA]
> mydf
    Month Pipe Gp Greater
 1:   Aug    3  A      NA
 2:   Oct    4  A    TRUE
 3:   Aug    5  B      NA
 4:   Oct    3  B   FALSE
 5:   Sep    2  B      NA
 6:   Aug    1  C      NA
 7:   Oct    3  D      NA
 8:   Aug    4  E      NA
 9:   Oct    3  E   FALSE
10:   Aug    5  F      NA
11:   Oct   NA  F      NA

---

您可以简化代码以避免以下两种情况：[.data.table如果需要，可以从上面调用，并避免定义管道this和管道last。

这可以通过两个联接来实现。第一个联接过滤出要比较的月份，并根据需要对其排序。然后可以进行比较。第二个联接将结果附加到原始数据帧

library(data.table)
# Last_Month_to_compare, This_month_to_compare
months_to_compare <- c("Aug", "Oct")
mDT <- setDT(mydf)[
  # append row id column (to preserve original order)
  , rn := .I][
    # cross join of groups and months
    CJ(Gp = Gp, Month = months_to_compare, unique = TRUE), on = .(Gp, Month)][
      # groupwise comparison of the two months
      , Greater := Pipe > shift(Pipe), by = Gp][]
# appending result to original data frame by joining with intermediate result
mydf[mDT, on = .(rn), Greater := i.Greater][]

请注意，mydf的原始顺序被保留

mDT的中间结果如下所示

编辑：补充说明 OP要求解释mydf[mDT，on=.rn]和mydf[mDT，on=.rn，morer:=i.morer][]之间的差异

对于data.table，X[Y，on=…]是一个右外部联接，它相当于mergeX，Y，all.Y=TRUE，即它返回Y的所有行，请参见

返回

以i.为前缀的列来自mDT。请注意，第6行和第7行在mydf中没有匹配的行。此外，行的顺序由mDT中的顺序决定

如果mydf和mDT互换

mDT[mydf, on = .(rn)][]

返回

以i.为前缀的列现在来自mydf。请注意，mDT中的第5行不匹配。此外，行的顺序由mydf确定

使用赋值运算符：=，X[Y，on=…，a:=b]将成为一个左内联接，它按原始顺序包含X的所有行。因此

mydf[mDT, on = .(rn), Greater := i.Greater][]

返回

---

其中，对于不匹配的行，较大值变为NA。

@Uwe-感谢您的帮助。您能帮助我理解mydf[mDT，on=.rn][]和mydf[mDT，on=.rn，较大值：=i.morer][]之间的区别吗.我了解前者的情况。这是rn上的左连接，但我不确定Greater:=I.Greater做什么。我感谢您的帮助。

    Month Pipe Gp rn Greater
 1:   Aug    3  A  1      NA
 2:   Oct    4  A  2    TRUE
 3:   Aug    5  B  3      NA
 4:   Oct    3  B  4   FALSE
 5:   Aug    1  C  6      NA
 6:   Oct   NA  C NA      NA
 7:   Aug   NA  D NA      NA
 8:   Oct    3  D  7      NA
 9:   Aug    4  E  9      NA
10:   Oct    3  E  8   FALSE
11:   Aug    5  F 11      NA
12:   Oct   NA  F 10      NA

mydf[mDT, on = .(rn)]

    Month Pipe Gp rn i.Month i.Pipe i.Gp Greater
 1:   Aug    3  A  1     Aug      3    A      NA
 2:   Oct    4  A  2     Oct      4    A    TRUE
 3:   Aug    5  B  3     Aug      5    B      NA
 4:   Oct    3  B  4     Oct      3    B   FALSE
 5:   Aug    1  C  6     Aug      1    C      NA
 6:    NA   NA NA NA     Oct     NA    C      NA
 7:    NA   NA NA NA     Aug     NA    D      NA
 8:   Oct    3  D  7     Oct      3    D      NA
 9:   Aug    4  E  9     Aug      4    E      NA
10:   Oct    3  E  8     Oct      3    E   FALSE
11:   Aug    5  F 11     Aug      5    F      NA
12:   Oct   NA  F 10     Oct     NA    F      NA

mDT[mydf, on = .(rn)][]

    Month Pipe Gp rn Greater i.Month i.Pipe i.Gp
 1:   Aug    3  A  1      NA     Aug      3    A
 2:   Oct    4  A  2    TRUE     Oct      4    A
 3:   Aug    5  B  3      NA     Aug      5    B
 4:   Oct    3  B  4   FALSE     Oct      3    B
 5:    NA   NA NA  5      NA     Sep      2    B
 6:   Aug    1  C  6      NA     Aug      1    C
 7:   Oct    3  D  7      NA     Oct      3    D
 8:   Oct    3  E  8   FALSE     Oct      3    E
 9:   Aug    4  E  9      NA     Aug      4    E
10:   Oct   NA  F 10      NA     Oct     NA    F
11:   Aug    5  F 11      NA     Aug      5    F

mydf[mDT, on = .(rn), Greater := i.Greater][]

    Month Pipe Gp rn Greater
 1:   Aug    3  A  1      NA
 2:   Oct    4  A  2    TRUE
 3:   Aug    5  B  3      NA
 4:   Oct    3  B  4   FALSE
 5:   Sep    2  B  5      NA
 6:   Aug    1  C  6      NA
 7:   Oct    3  D  7      NA
 8:   Oct    3  E  8   FALSE
 9:   Aug    4  E  9      NA
10:   Oct   NA  F 10      NA
11:   Aug    5  F 11      NA