计算分组意味着忽略每一行值

Question

我正在使用以下代码计算每个 class 的分组平均值。我需要每个 class 的平均值，将其放在每一行，但忽略公式中每一行的值（参见 expected_mean 列）。虽然该 DT 方法确实计算平均值，但它不会忽略每一行（请参阅第 value_mean 列）。

## create dataset
dataset <- data.frame(matrix(ncol = 2, nrow = 6))
colnames(dataset) <- c('class','value')
dataset$class <- c(rep('A',3),rep('B',3))
dataset$value <- 1:6

## convert to DT and aggregate
setDT(dataset)
dataset[, value_mean := mean(value), by=class]

## expected means (without itself)
dataset$expected_means <- c(2.5,2,1.5,5.5,5,4.5)

还有这个returns：

   class value value_mean expected_means
       A     1          2            2.5
       A     2          2            2.0
       A     3          2            1.5
       B     4          5            5.5
       B     5          5            5.0
       B     6          5            4.5

我需要计算每个 class 的平均值，每行放置一次，但忽略当前值。例如，对于第一行，而不是 (1+2+3)/3，它应该只做 (2+3)/2

Answer 1

肯定有比sapply更有效的方法，但你可以这样做：

setDT(dataset)[, value_mean := sapply(1:.N, function(x) mean(value[-x])), by = class]

输出：

   class value expected_means value_mean
1:     A     1            2.5        2.5
2:     A     2            2.0        2.0
3:     A     3            1.5        1.5
4:     B     4            5.5        5.5
5:     B     5            5.0        5.0
6:     B     6            4.5        4.5

Answer 2

您可以使用 sqldf:

library(sqldf)

dataset <- data.frame(class = rep(c("A", "B"), each = 3), 
                      value = 1:6, 
                      stringsAsFactors = FALSE)

result = sqldf('select d.*, 
                t.sum * 1.0 / (t.count * 1.0) as value_mean, 
                (t.sum - d.value)*1.0/ ((t.count - 1) * 1.0) as expected_means
                from dataset as d JOIN 
                 (select class, sum(value) as sum, count(*) as count 
                  from dataset 
                  group by class) as t 
                on d.class = t.class')

Answer 3

这是另一个选项：

dataset[, expected_means := (sum(value) - value) / (.N - 1L), class]

计算分组意味着忽略每一行值

Calculate grouped means ignoring each row value

r

mean