我有一个过去12小时内交易的加权总和的工作示例。现在，我添加了一个帐户列，并希望按组单独计算此加权和。代码将按如下所示运行。取消注解以# account开头的行，以将account列添加到df。如何修改倒数第二行代码，使其在每个account上分别计算rollapplyr？

library(zoo)
library(tidyverse)

创建示例数据：

set.seed(123)
randomDates <- function(N, st="2017-01-01 00:00:00", et="2017-02-01 23:59:59") {
     st <- as.POSIXct(st, tz = "UTC")

     et <- as.POSIXct(et, tz = "UTC")

     dt <- as.numeric(difftime(et,st,units="sec", tz="UTC"))
     ev <- sort(runif(N, 0, dt))
     rt <- st + ev
     rt
}

df <- data.frame(date = randomDates(100) , 
                 data = round( abs(rnorm(100)) * 100 ) # ,
                 # account = sample(c("A", "B", "C"), 100, replace=TRUE  )
)

df <- df %>% arrange(date)

定义帮助函数：

tau <- 0.00005
decay = function(tau, day){
  exp(-tau * day)
}

weighted <- function(x, tau) {
  tx <- as.numeric(time(x))
  seconds <- tail(tx, 1) - tx
  w <- (seconds < 43200) * decay(tau, seconds) # 12 hours in seconds
  sum(w * coredata(x))
}

计算滚动总和：

# Would like to modify this block to group by account
newData <- df %>%
  read.zoo  %>% 
  rollapplyr(43200, weighted, tau = tau, partial = TRUE, coredata = FALSE)

dfNew <- df %>% mutate( weighted_sum = newData  )

                   date data weighted_sum
1   2017-01-01 00:21:26   38    38.000000
2   2017-01-01 21:29:53   56    56.000000
3   2017-01-02 14:02:43   34    34.000000
4   2017-01-02 20:41:28    9    19.279179
5   2017-01-03 06:08:07  160   161.644215

根据我的研究，我还没有找到答案：
Apply a rolling sum by group in R
use rollapply and zoo to calculate rolling average of a column of variables
https://www.rdocumentation.org/packages/zoo/versions/1.8-1/topics/rollapply

我还根据对此问题的反馈和链接的可能重复的答案尝试了此解决方案。但是，应用相同的模式会导致无法解决的错误：

newData <- df %>%
  group_by(account) %>% 
  mutate(weighted_sum =  rollapplyr(., width=43200, FUN = weighted, 
                         tau = tau, partial = TRUE, coredata = FALSE)   ) %>% 
  ungroup()

引发此错误：

# Error in mutate_impl(.data, dots) : 
  Evaluation error: non-numeric argument to binary operator.