根据类别按日期计算累积产品

Question

我想在data.table中添加一个新列,其中包含Data1基于的累积产品Date.应为每个类别(Cat)计算累积产品,并应从最新的可用产品开始Date.

样本数据:

     DF = data.frame(Cat=rep(c("A","B"),each=4), Date=rep(c("01-08-2013","01-07-2013","01-04-2013","01-03-2013"),2), Data1=c(1:8))
     DF$Date = as.Date(DF$Date , "%m-%d-%Y")
     DT = data.table(DF)
     DT[ , Data1_cum:=NA_real_]
     DT

        Cat      Date Data1 Data1_cum
     1:  A 2013-01-08     1    NA
     2:  A 2013-01-07     2    NA
     3:  A 2013-01-04     3    NA
     4:  A 2013-01-03     4    NA
     5:  B 2013-01-08     5    NA
     6:  B 2013-01-07     6    NA
     7:  B 2013-01-04     7    NA
     8:  B 2013-01-03     8    NA

结果应如下所示:

        Cat      Date Data1 Data1_cum
     1:  A 2013-01-08     1    1
     2:  A 2013-01-07     2    2
     3:  A 2013-01-04     3    6
     4:  A 2013-01-03     4    24
     5:  B 2013-01-08     5    5
     6:  B 2013-01-07     6    30
     7:  B 2013-01-04     7    210
     8:  B 2013-01-03     8    1680

我发现我可以做类似的事情cumprod(),但我不知道如何处理类别.NAs in Data1应该被忽略/视为1.真正的数据集有大约800万行和1000个类别.

Answer 1

如果唯一的外观问题是订购......

DT[order(Date, decreasing=TRUE), Data1_cum := cumprod(Data1), by=Cat]
DT
   Cat       Date Data1 Data1_cum
1:   A 2013-01-08     1         1
2:   A 2013-01-07     2         2
3:   A 2013-01-04     3         6
4:   A 2013-01-03     4        24
5:   B 2013-01-08     5         5
6:   B 2013-01-07     6        30
7:   B 2013-01-04     7       210
8:   B 2013-01-03     8      1680

但是，如果您要处理 NA，则需要执行一些额外步骤：

注意：如果您打乱行的顺序，则结果可能会有所不同。小心执行order(.)命令的方式

  ## Let's add some NA values
  DT <- rbind(DT, DT)
  DT[c(2, 6, 11, 15), Data1 := NA]

  # shuffle the rows, to make sure this is right
  set.seed(1)
  DT <- DT[sample(nrow(DT))]

分配累积乘积：

离开 NA

## If you want to leave the NA's as NA's in the cum prod, use: 
DT[ , Data1_cum := NA_real_ ]
DT[ intersect(order(Date, decreasing=TRUE), which(!is.na(Data1))) 
      , Data1_cum := cumprod(Data1)
      , by=Cat]

# View the data, orderly
DT[order(Date, decreasing=TRUE)][order(Cat)]

     Cat       Date Data1 Data1_cum
  1:   A 2013-01-08     1         1
  2:   A 2013-01-08     1         1
  3:   A 2013-01-07     2         2
  4:   A 2013-01-07    NA        NA  <~~~~~~~  Note that the NA rows have the value of the prev row     
  5:   A 2013-01-04     3         6
  6:   A 2013-01-04    NA        NA  <~~~~~~~  Note that the NA rows have the value of the prev row
  7:   A 2013-01-03     4        24
  8:   A 2013-01-03     4        96
  9:   B 2013-01-08     5         5  
 10:   B 2013-01-08     5        25
 11:   B 2013-01-07     6       150
 12:   B 2013-01-07    NA        NA  <~~~~~~~  Note that the NA rows have the value of the prev row  
 13:   B 2013-01-04     7      1050
 14:   B 2013-01-04    NA        NA  <~~~~~~~  Note that the NA rows have the value of the prev row    
 15:   B 2013-01-03     8      8400
 16:   B 2013-01-03     8     67200

将 NA 替换为前一行的值

## If instead you want to treat the NA's as 1, use: 
DT[order(Date, decreasing=TRUE), Data1_cum := {Data1[is.na(Data1)] <- 1;  cumprod(Data1 [order(Date, decreasing=TRUE)] )}, by=Cat]

# View the data, orderly
DT[order(Date, decreasing=TRUE)][order(Cat)]

    Cat       Date Data1 Data1_cum
 1:   A 2013-01-08     1         1
 2:   A 2013-01-08     1         1
 3:   A 2013-01-07     2         2
 4:   A 2013-01-07    NA         2   <~~~~~~~ Rows with NA took on values of the previous Row
 5:   A 2013-01-04     3         6
 6:   A 2013-01-04    NA         6   <~~~~~~~ Rows with NA took on values of the previous Row
 7:   A 2013-01-03     4        24
 8:   A 2013-01-03     4        96
 9:   B 2013-01-08     5         5
10:   B 2013-01-08     5        25
11:   B 2013-01-07     6       150
12:   B 2013-01-07    NA       150   <~~~~~~~ Rows with NA took on values of the previous Row
13:   B 2013-01-04     7      1050
14:   B 2013-01-04    NA      1050   <~~~~~~~ Rows with NA took on values of the previous Row
15:   B 2013-01-03     8      8400
16:   B 2013-01-03     8     67200

或者，如果您已经有了累积乘积并且只想删除 NA，您可以按如下方式操作：

# fix the NA's with the previous value
DT[order(Date, decreasing=TRUE),
      Data1_cum := {tmp <- c(0, head(Data1_cum, -1));  
      Data1_cum[is.na(Data1_cum)] <- tmp[is.na(Data1_cum)]; 
      Data1_cum }
      , by=Cat ]