使用键来合并列，该键根据后缀R指定哪个列应该按优先级排序

Question

我有以下数据帧：

set.seed(123)
    df <- data.frame(col.dfA=rnorm(8,40,2),
                     col.dfB=rnorm(8,20,2),
                     colt=rnorm(8,100,20),
                     pr.dfA=rnorm(8,20,2),
                     pr.dfB=rnorm(8,30,2),
                     priority=c("dfA","dfB","dfA","dfA","dfB","dfA","dfB","dfB"))

现在我想分别合并col.dfA和col.dfB以及pr.dfA和pr.dfBinto df$col和df$pr。 df$priority列指示应在每行中使用哪些列值。 IE浏览器。在第1行，第3行，第4行和第6行中，应优先考虑带后缀dfA的列。在剩余行中，应优先考虑带有后缀dfB的列。该解决方案需要适用于具有数十列和数百行的较大场景。

初始数据帧：

   col.dfA  col.dfB      colt   pr.dfA   pr.dfB priority
1 38.87905 18.62629 109.95701 18.74992 31.79025      dfA
2 39.53965 19.10868  60.66766 16.62661 31.75627      dfB
3 43.11742 22.44816 114.02712 21.67557 31.64316      dfA
4 40.14102 20.71963  90.54417 20.30675 31.37728      dfA
5 40.25858 20.80154  78.64353 17.72373 31.10784      dfB
6 43.43013 20.22137  95.64050 22.50763 29.87618      dfA
7 40.92183 18.88832  79.47991 20.85293 29.38807      dfB
8 37.46988 23.57383  85.42218 19.40986 29.23906      dfB

预期结果：

  col      colt      pr            priority
1 38.87905 109.95701 18.74992      dfA
2 19.10868  60.66766 31.75627      dfB
3 43.11742 114.02712 21.67557      dfA
4 40.14102  90.54417 20.30675      dfA
5 20.80154  78.64353 31.10784      dfB
6 43.43013  95.64050 22.50763      dfA
7 18.88832  79.47991 29.38807      dfB
8 23.57383  85.42218 29.23906      dfB

Answer 1

如果你这样试试怎么办：

library(tidyverse)

df %>%
  gather(key = 'varname', value = 'varvalue', 
         matches(paste(.$priority, collapse = '|'))) %>%
  rowwise() %>%
  filter(grepl(priority, varname)) %>%
  mutate(varname = gsub('\\.[[:alpha:]]+$', '', varname)) %>%
  spread(key = varname, value = varvalue)

Answer 2

有一种替代方法，它使用data.table的melt()化身的能力同时重塑多个值列：

library(data.table)
library(magrittr)  # used to improve readability
# pick column names ending in .dfA or .dfB
pat <- names(df) %>% stringr::str_extract(".+\\.df(?=[AB]$)") %>% unique() %>% na.omit()
# new column names without trailing .df
col <- pat %>% stringr::str_replace("\\.df$", "")
# reshape from wide to  long format
melt(setDT(df)[, rn := .I], measure.vars = patterns(pat), value.name = col)[
  # subset rows where priority equals variable (as factor levels)
  as.integer(priority) == as.integer(variable)][
    # re-order and clean up
    order(rn)][, variable := NULL][]

        colt priority rn      col       pr
1: 109.95701      dfA  1 38.87905 18.74992
2:  60.66766      dfB  2 19.10868 31.75627
3: 114.02712      dfA  3 43.11742 21.67557
4:  90.54417      dfA  4 40.14102 20.30675
5:  78.64353      dfB  5 20.80154 31.10784
6:  95.64050      dfA  6 43.43013 22.50763
7:  79.47991      dfB  7 18.88832 29.38807
8:  85.42218      dfB  8 23.57383 29.23906

Answer 3

我不确定我明白你在追求什么......但我认为就是这样：

library(data.table)
setDT(df, key = 'priority')
df[.('dfA'), c('col', 'pr') := .(col.dfA, pr.dfA)]
df[.('dfB'), c('col', 'pr') := .(col.dfB, pr.dfB)]

如果需要，删除源列：

df[ , grep('\\.df[AB]$', names(df)) := NULL][]
#         colt priority      col       pr
# 1:  66.99376      dfA 40.81422 22.05758
# 2: 104.97938      dfA 41.09455 19.78682
# 3:  66.33725      dfA 40.08257 21.65990
# 4: 112.25391      dfA 42.89993 19.72432
# 5:  72.14824      dfB 23.47501 30.76038
# 6: 103.57393      dfB 19.43745 26.52382
# 7:  93.28548      dfB 20.46695 29.38035
# 8:  89.28845      dfB 19.21581 28.66537

如果您需要重新排序列，请使用setcolorder。

更具规划性的概括可以使用mget：

priorities = unique(df$priority)
nm = names(df)
merge_col = unique(c(sapply(priorities, function(suff) {
  has_suff = grepl(sprintf('%s$', suff), nm)
  gsub(sprintf('\\.%s$', suff), '', nm[has_suff])
})))

for (PRIORITY in priorities) {
  source_cols = sprintf('%s.%s', merge_cols, PRIORITY)
  df[.(PRIORITY), (merge_cols) := mget(source_cols)]
  df[ , (source_cols) := NULL]
}

众所周知，mget很慢。使用eval可能更有效：

df[.(PRIORITY), (merge_cols) := lapply(source_cols, function(x) eval(as.name(x)))]