以有效的方式在函数列表中查找各种分位数？

Question

我已经尝试多次调用分位数，但考虑到我需要 0.05 的分位数增量并且我的数据非常大，这非常低效。

ff <- function(table, IDs, q.increment=0.05){
  fun <- list("mean" = function(x) mean(x, na.rm = TRUE),
              "median" = function(x) median(x, na.rm =TRUE),
              function(x) quantile(x, probs = seq(0, 1, q.increment), na.rm=T))

  fun.names <- names(fun)
  abc <- table[ , c(summary = list(fun.names),
                    lapply(.SD, function(col) I(lapply(fun, function(f) f(col))))),
                by = IDs, .SDcols = grep("value", names(table))]
  return(abc)
}

dt <- data.table(country = c("FR", "US", "HU", "HU", "FR", "FR", "US", "US", "US", "HU"), value1=rnorm(10), value2=rnorm(10))
abc <- ff(dt, c("country"))

我不喜欢的是分位数都在一个单元格中，我希望每个分位数都有一行。

Answer 1

执行此操作的方法是将quantile 转换为list，然后unlisting 使用recursive = FALSE

dt[, c(unlist(lapply(.SD, function(x) list(mean = mean(x),
    median = median(x))), recursive = FALSE), 
   unlist(lapply(.SD, function(x) as.list(quantile(x,
   probs = seq(0, 1, 0.05)))), recursive = FALSE)), country]

---

它可以被包装在一个函数中

ff1 <- function(data, IDs, q.increment = 0.05) {
    f1 <- function(x) list(mean = mean(x, na.rm = TRUE),
                           median = median(x, na.rm = TRUE),
                           quantile = as.list(quantile(x, 
                probs = seq(0, 1, q.increment))))


     data[, unlist(unlist(lapply(.SD, f1), recursive = FALSE),
           recursive = FALSE), by = IDs, .SDcols = grep("value", names(data))]
 }

out <- ff1(dt, "country")

如果我们需要长格式，则使用melt

nm1 <- unique(sub(".*\\.", "", names(out)[-1]))
melt(out, measure = patterns('^value1', '^value2'),
      variable.name = 'summary')[, summary := nm1[summary]][]
# country summary      value1      value2
# 1:      FR    mean -0.70362861 -0.37004727
# 2:      US    mean -0.17024421 -0.10986835
# 3:      HU    mean  0.35754440  0.43067053
# 4:      FR  median -0.25453398 -0.72539656
# 5:      US  median -0.08068703  0.15472558
# 6:      HU  median  0.61732639  0.30846369
# 7:      FR      0% -1.60473855 -1.34258692
# 8:      US      0% -0.87641285 -2.04386860
# 9:      HU      0% -0.37871048  0.08147549
#10:      FR      5% -1.46971809 -1.28086789
#11:      US      5% -0.80765939 -1.72964937
#...