count number of times a factor appears during rolling window

问题

I want to generate the column: "PriorityCountInLast7Days". For a given employee A, this column counts the number of CASES in the last 7 days where PRIORITY is the same as the current case. How would I do that in R with the first 4 columns?

data <- data.frame(
    Date = c("2018-06-01", "2018-06-03", "2018-06-03", "2018-06-03",  "2018-06-04", "2018-06-01", "2018-06-02", "2018-06-03"),
Emp1 = c("A","A","A","A","A","A","B","B","B"),
Case = c("A1", "A2", "A3", "A4", "A5", "A6", "B1", "B2", "B3"),
Priority = c(0,0,0,1,2,0,0,0,0),
PriorityCountinLast7days = c(0,1,2,1,1,3,1,2,3))

+------------+------+------+----------+--------------------------+
|    Date    | Emp1 | Case | Priority | PriorityCountinLast7days |
+------------+------+------+----------+--------------------------+
| 2018-06-01 | A    | A1   |        0 |                        0 |
| 2018-06-03 | A    | A2   |        0 |                        1 |
| 2018-06-03 | A    | A3   |        0 |                        2 |
| 2018-06-03 | A    | A4   |        1 |                        1 |
| 2018-06-03 | A    | A5   |        2 |                        1 |
| 2018-06-04 | A    | A6   |        0 |                        3 |
| 2018-06-01 | B    | B1   |        0 |                        1 |
| 2018-06-02 | B    | B2   |        0 |                        2 |
| 2018-06-03 | B    | B3   |        0 |                        3 |
+------------+------+------+----------+--------------------------+

回答1:

You can accomplish this rolling window with an iterative conditional sum on your full dataset. What does this mean? Within a for loop you can check to see that your current date >= dates you want to include AND the dates you want to include >= to the date 7 days ago AND the cases you want to include are == to your current case. This logic combination in a loop will create this rolling filter for you. Here is a function:

rollPriority <- function(data, window = 7){
  stopifnot(all(c("Date","Case","Priority") %in% colnames(data))) # string error check
  data$Date <- as.Date(data$Date)
  for(i in 1:nrow(data)){
    #priorxdays <= dates we want <= current date
    datecheck <- (data$Date[i] - (window-1)) <= data$Date & data$Date <= data$Date[i]
    casecheck <- data$Case == data$Case[i]
    data$PriorityCountinLastXdays[i] = sum(data$Priority[which(datecheck & casecheck)])
  }
  Xdays <- which(colnames(data) == "PriorityCountinLastXdays")
  colnames(data)[Xdays] <- paste0("PriorityCountinLast", window, "days")
  return(data)
}

In the future, please provide example data with a reproducible output. You will notice that we cannot match your expected 7 day rolling output having only seen 4 days of information. A quick method here is to use expand.grid() to generate combinations, and set.seed() to preserve sampling output:

# Reproducible Example Data
dat <- expand.grid(Date = seq.Date(as.Date("2018-06-01"),
                                   as.Date("2018-06-4"), 
                                   by = "day"), 
                   Case = as.factor(sort(apply(expand.grid(c("A","B"),1:2), 
                                               1, 
                                               paste0, 
                                               collapse = ""))))
# Ensures random sampling is identical each time
set.seed(42); 
dat$Priority <- sample(0:1, nrow(dat), replace = T)

# The function
rollPriority(dat, 2)
#         Date Case Priority PriorityCountinLast2days
#1  2018-06-01   A1        1                        1
#2  2018-06-02   A1        1                        2
#3  2018-06-03   A1        0                        1
#4  2018-06-04   A1        1                        1
#5  2018-06-01   A2        1                        1
#6  2018-06-02   A2        1                        2
#7  2018-06-03   A2        1                        2
#8  2018-06-04   A2        0                        1
#9  2018-06-01   B1        1                        1
#10 2018-06-02   B1        1                        2
#11 2018-06-03   B1        0                        1
#12 2018-06-04   B1        1                        1
#13 2018-06-01   B2        1                        1
#14 2018-06-02   B2        0                        1
#15 2018-06-03   B2        0                        0
#16 2018-06-04   B2        1                        1

This way it is easier for someone to accurately assist you.

来源：https://stackoverflow.com/questions/52023433/count-number-of-times-a-factor-appears-during-rolling-window

标签

date

count

levels

rolling-computation