在熔化时保留变量级别的名称

Question

这个问题在这里已经有了答案:





Melt using patterns when variable names contain string information - avoid coercion to numeric

(1 个回答)


3年前关闭。




有没有办法保留正在融化的变量的原始级别的名称？例如，在下面的例子中，有没有办法得到“alpha”、“beta”和“gamma”而不是“1”、“2”、“3”。

当然，我可以重命名它们，但我正在使用的数据集有大量级别，因此重命名它们将非常耗时且容易出错。

谢谢。

library(data.table)
#> Warning: package 'data.table' was built under R version 3.4.2
set.seed(2334)

# define the dataframe
df <-
  as.data.frame(
    cbind(
      a_alpha = rnorm(10),
      a_beta = rnorm(10),
      a_gamma = rnorm(10),
      b_alpha = rnorm(10),
      b_beta = rnorm(10),
      b_gamma = rnorm(10),
      id = c(1:10)
    )
  )

# check the structure of the wide format data
str(df)
#> 'data.frame':    10 obs. of  7 variables:
#>  $ a_alpha: num  -0.118 1.237 0.809 -0.766 -0.592 ...
#>  $ a_beta : num  0.0019 1.0639 2.336 0.9056 0.6449 ...
#>  $ a_gamma: num  0.5485 0.8345 -0.5977 0.0827 0.2754 ...
#>  $ b_alpha: num  0.209 -0.305 0.434 -0.362 0.412 ...
#>  $ b_beta : num  -1.6404 2.8382 0.0661 0.7249 -0.4421 ...
#>  $ b_gamma: num  -0.144 0.964 -0.763 -1.356 0.995 ...
#>  $ id     : num  1 2 3 4 5 6 7 8 9 10

# convert to long format
df_long <- data.table::melt(
  data.table::setDT(df),
  measure = patterns("^a_", "^b_"),
  value.name = c("a", "b"),
  variable.name = "item"
)

# check the structure of the long format data
str(df_long)
#> Classes 'data.table' and 'data.frame':   30 obs. of  4 variables:
#>  $ id  : num  1 2 3 4 5 6 7 8 9 10 ...
#>  $ item: Factor w/ 3 levels "1","2","3": 1 1 1 1 1 1 1 1 1 1 ...
#>  $ a   : num  -0.118 1.237 0.809 -0.766 -0.592 ...
#>  $ b   : num  0.209 -0.305 0.434 -0.362 0.412 ...
#>  - attr(*, ".internal.selfref")=<externalptr>

# structure of item
levels(df_long$item)
#> [1] "1" "2" "3"

# Question: instead of "1" "2" "3", how to get the "item" factor levels to be: "alpha" "beta" "gamma"

创建于 2018-01-12 由 reprexpackage (v0.1.1.9000)。

Answer 1

我过去处理这个问题的方法是使用 factor后 melt数据。但是，您可能必须进行一些检查以确保数据和级别的顺序正确。

下面是一个例子:

set.seed(2334)
df <- data.table(a_alpha = rnorm(10), a_beta = rnorm(10), a_gamma = rnorm(10),
                 b_alpha = rnorm(10), b_beta = rnorm(10), b_gamma = rnorm(10), 
                 id = c(1:10))
df_mess <- copy(df)
setcolorder(df_mess, c(1, 7, 6, 4, 2, 5, 3))
names(df_mess)
# [1] "b_alpha" "id"      "a_alpha" "a_beta"  "b_beta"  "b_gamma" "a_gamma"

stubs <- c("^a_", "^b_")
## assumes all stubs have same number of cols. Easy to modify
labs <- grep(stubs[1], names(df_mess), value = TRUE) 
labs <- gsub(paste(stubs, collapse = "|"), "", labs[order(labs)])

out1 <- melt(df, measure.vars = patterns(stubs), value.name = c("a", "b"), 
             variable.name = "item")[
               , item := factor(item, labels = labs)][]

out2a <- melt(df_mess, measure.vars = patterns(stubs), value.name = c("a", "b"), 
              variable.name = "item")[
                , item := factor(item, labels = labs)][]

out2b <- melt(setcolorder(df_mess, names(df_mess)[order(names(df_mess))]),
     measure.vars = patterns(stubs), value.name = c("a", "b"), 
     variable.name = "item")[
       , item := factor(item, labels = labs)][]

library(compare)
compare(out1, out2a)
# FALSE [TRUE, TRUE, TRUE, FALSE]
compare(out1, out2b)
# TRUE

我没有做足够的测试用例来自信地说是否使用 order在 names和 levs对所有情况都足够了，但到目前为止，我还没有发现数据不平衡时的异常(exception)情况。