Question

我想通过将负值添加到每个组中的前一行，将当前行的负值转移到前一行。以下是我的原始数据示例：

raw_data <- data.frame(GROUP = rep(c('A','B','C'),each = 6),
                   YEARMO = rep(c(201801:201806),3),
                   VALUE = c(100,-10,20,70,-50,30,20,60,40,-20,-10,50,0,10,-30,50,100,-100))
> raw_data
  GROUP YEARMO VALUE
1      A 201801   100  
2      A 201802   -10
3      A 201803    20
4      A 201804    70
5      A 201805   -50
6      A 201806    30
7      B 201801    20
8      B 201802    60
9      B 201803    40
10     B 201804   -20
11     B 201805   -10
12     B 201806    50
13     C 201801     0
14     C 201802    10
15     C 201803   -30
16     C 201804    50
17     C 201805   100
18     C 201806  -100

以下是我想要的输出：

final_data <- data.frame(GROUP = rep(c('A','B','C'),each = 6),
                   YEARMO = rep(c(201801:201806),3),
                   VALUE = c(90,0,20,20,0,30,20,60,10,0,0,50,-20,0,0,50,0,0))
> final_data
   GROUP YEARMO VALUE
1      A 201801    90
2      A 201802     0
3      A 201803    20
4      A 201804    20
5      A 201805     0
6      A 201806    30
7      B 201801    20
8      B 201802    60
9      B 201803    10
10     B 201804     0
11     B 201805     0
12     B 201806    50
13     C 201801   -20
14     C 201802     0
15     C 201803     0
16     C 201804    50
17     C 201805     0
18     C 201806     0

以下数据帧将显示如何在每个组中进行转换：

Trans_GRP_A <- data.frame(GROUP = rep('A',each = 6),
                   YEARMO = c(201801:201806),
                   VALUE = c(100,-10,20,70,-50,30),
                   ITER_1 = c(100,-10,20,20,0,30),
                   ITER_2 = c(90,0,20,20,0,30))
> Trans_GRP_A
  GROUP YEARMO VALUE ITER_1 ITER_2
1     A 201801   100    100     90
2     A 201802   -10    -10      0
3     A 201803    20     20     20
4     A 201804    70     20     20
5     A 201805   -50      0      0
6     A 201806    30     30     30

> Trans_GRP_B <- data.frame(GROUP = rep('B',each = 6),
+                           YEARMO = c(201801:201806),
+                           VALUE = c(20,60,40,-20,-10,50),
+                           ITER_1 = c(20,60,40,-30,0,50),
+                           ITER_2 = c(20,60,10,0,0,50))
> Trans_GRP_B
  GROUP YEARMO VALUE ITER_1 ITER_2
1     B 201801    20     20     20
2     B 201802    60     60     60
3     B 201803    40     40     10
4     B 201804   -20    -30      0
5     B 201805   -10      0      0
6     B 201806    50     50     50

> Trans_GRP_C <- data.frame(GROUP = rep('C',each = 6),
+                           YEARMO = c(201801:201806),
+                           VALUE = c(0,10,-30,50,100,-100),
+                           ITER_1 = c(0,10,-30,50,0,0),
+                           ITER_2 = c(0,-20,0,50,0,0),
+                           ITER_3 = c(-20,0,0,50,0,0))
> Trans_GRP_C
  GROUP YEARMO VALUE ITER_1 ITER_2 ITER_3
1     C 201801     0      0      0    -20
2     C 201802    10     10    -20      0
3     C 201803   -30    -30      0      0
4     C 201804    50     50     50     50
5     C 201805   100      0      0      0
6     C 201806  -100      0      0      0

传输逻辑如下：

将负值替换为0。
将当前行的负值添加到上一行的值。
将负值传输到上一行，直到该值变为正值或0。
如果传输未产生正值，则传输直到组中的第一行，这里每个组中的第一行是YEARMO = 201801。

欢迎任何解决方案。我认为矢量化解决方案的执行速度可能会更快。

Answer 1

那是一个棘手的问题。我试图找到一种矢量化的解决方案，但到目前为止唯一有效的方法是在每个组中的行中循环遍历向后：

library(data.table)
DT <- as.data.table(raw_data)
DT$final <- final_data$VALUE
DT[, new := {
  x <- VALUE
  sn <- 0
  for (i in .N:1) {
    if (i > 1) {
      if (x[i] < 0) {
        sn <- sn + x[i]
        x[i] <- 0
      } else {
        tmp <- pmax(x[i] + sn, 0)
        sn <- sn + x[i] - tmp
        x[i] <- tmp
      }
    } else {
      x[i] <- x[i] + sn
    }
  }
  x
}, by = GROUP]
DT[]

    GROUP YEARMO VALUE final new
 1:     A 201801   100    90  90
 2:     A 201802   -10     0   0
 3:     A 201803    20    20  20
 4:     A 201804    70    20  20
 5:     A 201805   -50     0   0
 6:     A 201806    30    30  30
 7:     B 201801    20    20  20
 8:     B 201802    60    60  60
 9:     B 201803    40    10  10
10:     B 201804   -20     0   0
11:     B 201805   -10     0   0
12:     B 201806    50    50  50
13:     C 201801     0   -20 -20
14:     C 201802    10     0   0
15:     C 201803   -30     0   0
16:     C 201804    50    50  50
17:     C 201805   100     0   0
18:     C 201806  -100     0   0

sn存储，即累积负值，然后通过后续（反向顺序）正值“消耗”负值。

Answer 2

这里是另一种选择，将向量的正部分与向量的负部分递归求和，直到不再有负值或已执行了.N次（其中.N是行的行数每个组）

setDT(raw_data)[, OUTPUT := {
        posVal <- replace(VALUE, VALUE<0, 0)
        negVal <- replace(VALUE, VALUE>0, 0)
        n <- 1L
        while (any(negVal < 0) && n < .N) {
            posVal <- replace(posVal, posVal<0, 0) + 
                shift(negVal, 1L, type="lead", fill=0) +
                c(negVal[1L], rep(0, .N-1L))
            negVal <- replace(posVal, posVal>0, 0)
            n <- n + 1L
        }
        posVal
    }, by=.(GROUP)]

输出：

    GROUP YEARMO VALUE OUTPUT
 1:     A 201801   100     90
 2:     A 201802   -10      0
 3:     A 201803    20     20
 4:     A 201804    70     20
 5:     A 201805   -50      0
 6:     A 201806    30     30
 7:     B 201801    20     20
 8:     B 201802    60     60
 9:     B 201803    40     10
10:     B 201804   -20      0
11:     B 201805   -10      0
12:     B 201806    50     50
13:     C 201801     0    -20
14:     C 201802    10      0
15:     C 201803   -30      0
16:     C 201804    50     50
17:     C 201805   100      0
18:     C 201806  -100      0

如何将当前行中的负值传输到数据帧中的前一行？

2 个答案: