How to reorder factor of a subset of data.frame and apply it to main data.frame?

Question

I have a data.frame with two variables and a factor column. Then I compute a subset of this data.frame and want to reorder the remaining factors. I found the solution below. But it's to slow with real numbers. So how do I reorder my factor?

Here's a step-by-step example:

     library(plyr)
     library(ggplot2)
     # generate an example data.frame
     # x and y are integers, l is a factor
     df <- data.frame(x=rep(1:5, each=4), y=seq(1:5), l=factor(letters[seq( from = 1, to = 10 )]))
     df <- df[seq(1:17),]
     df
       x y l
    1  1 1 a
    2  1 2 b
    3  1 3 c
    4  1 4 d
    5  2 5 e
    6  2 1 f
    7  2 2 g
    8  2 3 h
    9  3 4 i
    10 3 5 j
    11 3 1 a
    12 3 2 b
    13 4 3 c
    14 4 4 d
    15 4 5 e
    16 4 1 f
    17 5 2 g

Now I compute a temporary data.frame which I will use to select a subset of df:

     # computing temporary data.frame
      df2 <- ddply(df, .(l), summarize, sum=sum(y))
      df2$pct <- df2$sum / sum(df2$sum) * 100
      df2
       l sum       pct
    1  a   2  4.166667
    2  b   4  8.333333
    3  c   6 12.500000
    4  d   8 16.666667
    5  e  10 20.833333
    6  f   2  4.166667
    7  g   4  8.333333
    8  h   3  6.250000
    9  i   4  8.333333
    10 j   5 10.416667
     # select only those letters with "high enough" y-value
     df2.selected <- df2[df2$pct > 10,]
     df2.selected
       l sum      pct
    3  c   6 12.50000
    4  d   8 16.66667
    5  e  10 20.83333
    10 j   5 10.41667
     # use only those letters which occur in df2.selected$l
     df.subset <- df[df$l %in% df2.selected$l,]
     df.subset
       x y l
    3  1 3 c
    4  1 4 d
    5  2 5 e
    10 3 5 j
    13 4 3 c
    14 4 4 d
    15 4 5 e

I get rid of the now unused values of my factor:

     # get rid of unused values of l
      df.subset$l <- factor(df.subset$l)
      str(df.subset)
    'data.frame': 7 obs. of  3 variables:
     $ x: int  1 1 2 3 4 4 4
     $ y: int  3 4 5 5 3 4 5
     $ l: Factor w/ 4 levels "c","d","e","j": 1 2 3 4 1 2 3

The new order of my subset-facotr should be this one (I need this for facet_wrap below):

      # the new order of the factor variable should be the (inverse) order of sum
      df2.selected <- df2.selected[order(-df2.selected$sum),]
      df2.selected
       l sum      pct
    5  e  10 20.83333
    4  d   8 16.66667
    3  c   6 12.50000
    10 j   5 10.41667
     # that should be the new order of the factor variable l: e, d, c, j
     # get rid of unused values of l
     df2.selected$l <- factor(df2.selected$l)
     df2.selected
       l sum      pct
    5  e  10 20.83333
    4  d   8 16.66667
    3  c   6 12.50000
    10 j   5 10.41667
     str(df2.selected)
    'data.frame': 4 obs. of  3 variables:
     $ l  : Factor w/ 4 levels "c","d","e","j": 3 2 1 4
     $ sum: int  10 8 6 5
     $ pct: num  20.8 16.7 12.5 10.4


      # Here I need the order e, f, c, j!
      ggplot(data=df.subset, aes(x=x, y=y)) + geom_point() + facet_wrap(~l)
      # so merged both -- This is the problem. It's too expensive. Is there a better way?
      df.merged <- merge(df.subset, df2.selected, by=c('l'))
      df.merged$l <- reorder(df.merged$l, -df.merged$sum)
      df.merged
      l x y sum      pct
    1 c 1 3   6 12.50000
    2 c 4 3   6 12.50000
    3 d 1 4   8 16.66667
    4 d 4 4   8 16.66667
    5 e 2 5  10 20.83333
    6 e 4 5  10 20.83333
    7 j 3 5   5 10.41667
     str(df.merged)
    'data.frame': 7 obs. of  5 variables:
     $ l  : Factor w/ 4 levels "e","d","c","j": 3 3 2 2 1 1 4
      ..- attr(*, "scores")= num [1:4(1d)] -6 -8 -10 -5
      .. ..- attr(*, "dimnames")=List of 1
      .. .. ..$ : chr  "c" "d" "e" "j"
     $ x  : int  1 4 1 4 2 4 3
     $ y  : int  3 3 4 4 5 5 5
     $ sum: int  6 6 8 8 10 10 5
     $ pct: num  12.5 12.5 16.7 16.7 20.8 ...
      ggplot(data=df.merged, aes(x=x, y=y)) + geom_point() + facet_wrap(~l)

Answer 1

Here is a solution with data.table that should be relatively fast:

library(data.table)
dt <- data.table(df, key="l")
keep.lvls <- as.character(
  dt[, list(sum=sum(y)), by=l][,    # get the sums for each group
    pct:=sum/sum(sum) * 100][       # pct for each group   
    pct > 10][                      # only keep those greater than 10 
    order(pct, decreasing=T), l]    # order by pct, pull out `l` only
)
str(dt.final <- 
  dt[
    keep.lvls,][,                     # only keep `keep.lvls` from `dt`
    l:=factor(l, levels=keep.lvls)])  # reset factors on `dt` to have `keep.lvls` levels

which produces:

Classes ‘data.table’ and 'data.frame':  8 obs. of  3 variables:
 $ l: Factor w/ 4 levels "e","j","d","i": 1 1 2 2 3 3 4 4
 $ x: int  2 4 3 5 1 4 3 5
 $ y: int  5 5 5 5 4 4 4 4
 - attr(*, ".internal.selfref")=<externalptr> 

Note these answers are slightly different to yours since we have different random data. This is with set.seed(1).