问题
I'd like to know how to make multiple sampling in R. For example, when I try dividing some data into 60(train data):40(validate data), I can write the code like this:
original.data = read.csv("~.csv", na.strings="")
train.index = sample(c(1:dim(original.data)[1]), dim(original.data)[1]*0.6)
train.data = original.data[train.index,]
valid.data = original.data[-train.index,]
However, it is so hard to figure out making multiple sampling like dividing some data into 60:20:20.
I would appreciate if you make me know the best solution!
回答1:
If you want more than two sets, then the other solutions are close but you need just a little more. There are at least two options.
First:
set.seed(2)
table(samp <- sample(1:3, size = nrow(iris), prob = c(0.6, 0.2, 0.2), replace = TRUE))
# 1 2 3
# 93 35 22
nrow(iris) # 150
set1 <- iris[samp == 1,]
set2 <- iris[samp == 2,]
set3 <- iris[samp == 3,]
set1 <- iris[samp == 1,]
set2 <- iris[samp == 2,]
set3 <- iris[samp == 3,]
nrow(set1)
# [1] 93
nrow(set2)
# [1] 35
nrow(set3)
# [1] 22
Because it's random, you want always get your exact proportions.
Second:
If you must have exact proportions, you can do this:
ns <- nrow(iris) * c(0.6, 0.2, 0.2)
sum(ns)
# [1] 150
### in case of rounding (and sum != nrow) ... just fix one of ns
rep(1:3, times = ns)
# [1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
# [46] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
# [91] 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
# [136] 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
set.seed(2)
head(samp <- sample(rep(1:3, times = ns)))
# [1] 1 2 1 1 3 3
set1 <- iris[samp == 1,]
set2 <- iris[samp == 2,]
set3 <- iris[samp == 3,]
nrow(set1)
# [1] 90
nrow(set2)
# [1] 30
nrow(set3)
# [1] 30
This can easily be generalized to support an arbitrary number of partitions.
来源:https://stackoverflow.com/questions/56277391/multiple-sampling-in-r