Question

我希望您能想到一种更优雅的方法，以计算出前几天发生的事件数量。我的代码（如下）可以工作，但是不是很好，也不是可伸缩的。我正在尝试到达底部的表（desired_table）。有什么想法吗？

我想以一种更为简洁的方式来计算前几天的事件总数。

require(data.table)

# simulating an example data.table
date <- c("2000-01-01", "2000-01-04", "2000-01-05", "2000-01-06", "2000-01-01", "2000-01-02", "2000-01-03", "2000-01-04", "2000-01-05", "2000-01-06" , "2000-01-01", "2000-01-04", "2000-01-05", "2000-01-06", "2000-01-01", "2000-01-02", "2000-01-03", "2000-01-04", "2000-01-05", "2000-01-06")

cohort <- c("a", "b", "c")

zz <- data.table(DATE = date, COHORT = cohort)
zz$DATE <- as.Date(zz$DATE)  # making sure the date is in the correct format

# adding on some other date fields so we can summarise by these days as well
zz$d1 <- zz$DATE +1  # will become "yesterday" when joined
zz$d2 <- zz$DATE +2  # will become "day before yesterday", when joined 

# summarising the data for the first date
summary1 <- zz[,list(events_today = .N ), by= c("DATE", "COHORT")]

# summarising the data for the previous
summary2 <- zz[,list(events_yesterday  = .N ), by= c("d1", "COHORT")]

# summarising the data for the day before yesterday
summary3 <- zz[,list(events_day_before_yesterday  = .N ), by= c("d2", "COHORT")]

# merging the tables together
summary1.2 <- merge(summary1, summary2, by.x = c("DATE", "COHORT"), by.y = c("d1", "COHORT"), all = TRUE)

# merging the tables together to join on third summary table.
desired_table <- merge(summary1.2, summary3, by.x = c("DATE", "COHORT"), by.y = c("d2", "COHORT"), all = TRUE)

print(desired_table)

必须要有一种更优雅的方法吗？
我的示例很简单-实际上，我可能想对成千上万条记录和许多时间段进行此操作。

Answer 1

我认为一种更优雅的方式

long_zz <- melt(zz, id.vars = "COHORT")
new_zz <- dcast(long_zz, COHORT + value ~ variable, fun = length, drop = FALSE, fill = NA)
new_zz
#     COHORT      value DATE d1 d2
#  1:      a 2000-01-01    1 NA NA
#  2:      a 2000-01-02    1  1 NA
#  3:      a 2000-01-03    1  1  1
#  4:      a 2000-01-04   NA  1  1
#  5:      a 2000-01-05    2 NA  1
#  6:      a 2000-01-06    2  2 NA
#  7:      a 2000-01-07   NA  2  2
#  8:      a 2000-01-08   NA NA  2
#  9:      b 2000-01-01    2 NA NA
# 10:      b 2000-01-02   NA  2 NA
# 11:      b 2000-01-03    1 NA  2
# 12:      b 2000-01-04    2  1 NA
# 13:      b 2000-01-05   NA  2  1
# 14:      b 2000-01-06    2 NA  2
# 15:      b 2000-01-07   NA  2 NA
# 16:      b 2000-01-08   NA NA  2
# 17:      c 2000-01-01    1 NA NA
# 18:      c 2000-01-02    1  1 NA
# 19:      c 2000-01-03   NA  1  1
# 20:      c 2000-01-04    2 NA  1
# 21:      c 2000-01-05    2  2 NA
# 22:      c 2000-01-06   NA  2  2
# 23:      c 2000-01-07   NA NA  2
# 24:      c 2000-01-08   NA NA NA
#     COHORT      value DATE d1 d2

在这里，我首先将数据从宽格式转换为长格式，然后使用种姓将变量（DATE，d1，d2）再次拆分为列，同时计算每个COHORT值组中的行数每个变量。如果没有drop = FALSE，我们将错过第24行，在该行中COHORT c不会发生任何事件。

您可以使用

设置名称

setnames(new_zz, c("value", "DATE", "d1", "d2"), c("DATE", "events_today","events_yesterday","events_day_before_yesterday"))

mircobenchmark-您的方法（合并）与我的方法（long_wide）的结果：

Unit: milliseconds
      expr       min        lq     mean    median        uq      max neval
     merge 14.740983 18.534281 31.88430 21.223305 31.830966 353.3662   100
 long_wide  5.102077  6.411999 10.82941  7.130821  8.884161 117.7351   100

您是否知道一种更优雅的方式来计算前几天的事件数？

1 个答案: