在R中创建3行的随机组

Question

I'm trying to create as many random groups from a dataset as possible. My data is kind of complicated to explain so I'll use iris for my example.

In iris , the Species variable contains 3 unique values: setosa , versicolor , and virginica .

1. I want to randomize and group the dataset into groups of 3 rows, with each group containing unique Species only. (eg. 1 of each Species)

2. Each group must have a cumsum(Sepal.Width >= 10)

3. Create a new ID that identifies each group.

So far I've tried using the dplyr function group_by() and sample_n() . Also split() and sample() , but can't seem to get the desired result.

Using split() I think might be the wrong way to do it. I was trying to make it work along these lines with no luck.

split(unique(iris), sample(1:nrow(iris) %/% 3))

Answer 1

Try something like this:

#the sample
N=dim(iris)[1]
n=50 #sample size
set.seed(123)
si=iris[sample(N,n),c("Species","Sepal.Width")]

#the "cumsum"
lim=2.8 #for the conditional sum
Sepal.Width=sapply(split(si,si$Species),function(x)
   sum(x$Sepal.Width >= lim))
sol=data.frame(Species=names(Sepal.Width),Sepal.Width)
sol$ID=1:length(sol[,1])
sol
#               Species Sepal.Width ID
# setosa         setosa          18  1
# versicolor versicolor           8  2
# virginica   virginica          14  3

Answer 2

I think I understood the problem. Here's how you could do it using dplyr

First, load some packages and add a unique ID for each row in the iris data.frame.

library(dplyr)
library(tidyr)
iris = iris %>% mutate(Row.ID=1:n())

Then, let's split the Row.IDs according to species, and get a data.frame with all possible combinations of one row from each species

iris_split = split(iris$Row.ID, iris$Species)
combinations = do.call(expand.grid, iris_split)

Now, it's dplyr and tidyr time. Let's gather those combinations in a variable called tmp , join tmp with the rest of the iris data.frame and then filter according to the criteria.

tmp = combinations %>%
    mutate(Group.ID=1:n()) %>%
    gather(Var, Row.ID, -Group.ID) %>%
    select(-Var)

result = iris %>%
    inner_join(tmp) %>%
    group_by(Group.ID) %>%
    filter(sum(Sepal.Length) > 10) %>%
    arrange(Group.ID)

The result data.frame should be what you're looking for.