Conditional filtering based on the level of a fact

问题:

I would like to clean up the following code. Specifically, I'm wondering if I can consolidate the three filter statements so that I end up with the final data.frame (the rind()) that contains the row of data "spring" if it exists, the row of data for "fall" if "spring" doesn't exist, and finally the row of data if neither "spring" nor "fall" exist. The code below seems very clunky and inefficient. I am trying to free myself of for(), so hopefully the solution won't involve one. Could this be done using dplyr?

# define a %not% to be the opposite of %in%
library(dplyr)
`%not%` <- Negate(`%in%`)
f <- c("a","a","a","b","b","c")
s <- c("fall","spring","other", "fall", "other", "other")
v <- c(3,5,1,4,5,2)
(dat0 <- data.frame(f, s, v))
sp.tmp <- filter(dat0, s == "spring")
fl.tmp <- filter(dat0, f %not% sp.tmp$f, s == "fall")
ot.tmp <- filter(dat0, f %not% sp.tmp$f, f %not% fl.tmp$f, s == "other")
rbind(sp.tmp,fl.tmp,ot.tmp)

回答1:

It looks like within each group of f, you want to extract the row of, in descending order of preference, spring, fall, or other.

If you first make your ordering of preference the actual factor ordering:

dat0$s <- factor(dat0$s, levels=c("spring", "fall", "other"))

Then you can use this dplyr solution to get the minimum row (relative to that factor) within each group:

newdat <- dat0 %.% group_by(f) %.% filter(rank(s) == 1)