R data.table - 将函数 A 应用于某些列，将函数 B 应用于其他一些列

Question

我想聚合数据表的行，但是聚合函数取决于列的名称。

例如，如果列名是：

• variable1 或 variable2，然后应用 mean() 函数。
• variable3，然后应用 max() 函数。
• variable4，然后应用 sd() 函数。

我的数据表总是有一个datetime 列：我想按时间聚合行。 但是，“数据”列的数量可能会有所不同。

我知道如何对所有列使用相同的聚合函数（例如 mean()）：

dt <- dt[, lapply(.SD, mean),
           by = .(datetime = floor_date(datetime, timeStep))]

或仅用于列的子集：

cols <- c("variable1", "variable2")    
dt <- dt[ ,(cols) := lapply(.SD, mean), 
            by = .(datetime = floor_date(datetime, timeStep)),
            .SDcols = cols]

我想做的是这样的：

colsToMean <- c("variable1", "variable2") 
colsToMax <- c("variable3")   
colsToSd <- c("variable4")   
dt <- dt[ ,{(colsToMean) := lapply(.SD???, mean),
             (colsToMax) := lapply(.SD???, max),
             (colsToSd) :=  lapply(.SD???, sd)}, 
            by = .(datetime = floor_date(datetime, timeStep)),
            .SDcols = (colsToMean, colsToMax, colsToSd)]

我查看了data.table in R - apply multiple functions to multiple columns，它给了我使用自定义函数的想法：

myAggregate <- function(x, columnName) {
   FUN = getAggregateFunction(columnName) # Return mean() or max() or sd()
   return FUN(x)
}
dt <- dt[, lapply(.SD, myAggregate, ???columName???),
           by = .(datetime = floor_date(datetime, timeStep))]

但我不知道如何将当前列名传递给myAggregate()...

Answer 1

这是使用Map 或mapply 的一种方法：

让我们先制作一些玩具数据：

dt <- data.table(
    variable1 = rnorm(100),
    variable2 = rnorm(100),
    variable3 = rnorm(100),
    variable4 = rnorm(100),
    grp = sample(letters[1:5], 100, replace = T)
)

colsToMean <- c("variable1", "variable2") 
colsToMax <- c("variable3")   
colsToSd <- c("variable4")

那么，

scols <- list(colsToMean, colsToMax, colsToSd)
funs <- rep(c(mean, max, sd), lengths(scols))

# summary
dt[, Map(function(f, x) f(x), funs, .SD), by = grp, .SDcols = unlist(scols)]

# or replace the original values with summary statistics as in OP
dt[, unlist(scols) := Map(function(f, x) f(x), funs, .SD), by = grp, .SDcols = unlist(scols)]

启用 GForce 的另一个选项：

scols <- list(colsToMean, colsToMax, colsToSd)
funs <- rep(c('mean', 'max', 'sd'), lengths(scols))

jexp <- paste0('list(', paste0(funs, '(', unlist(scols), ')', collapse = ', '), ')')
dt[, eval(parse(text = jexp)), by = grp, verbose = TRUE]

# Detected that j uses these columns: variable1,variable2,variable3,variable4 
# Finding groups using forderv ... 0.000sec 
# Finding group sizes from the positions (can be avoided to save RAM) ... 0.000sec 
# Getting back original order ... 0.000sec 
# lapply optimization is on, j unchanged as 'list(mean(variable1), mean(variable2), max(variable3), sd(variable4))'
# GForce optimized j to 'list(gmean(variable1), gmean(variable2), gmax(variable3), gsd(variable4))'
# Making each group and running j (GForce TRUE) ... 0.000sec