按条件分组的样本值

一个解决方案是创建一个独特的组标签列表,并将其洗牌(这里我通过 nest )。然后,根据组是否位于数据帧的前25%行中,您可以指定a)介于1和2之间的随机数,或b)始终为2。最后,您可以使用分配的编号定义如何对每个组的0和1进行采样,然后取消对结果的测试。

set.seed(0)
result <- df %>% 
  nest(data = -group) %>% 
  .[sample(1:nrow(.), nrow(.)), ] %>% # shuffle the group order
  mutate(
    value_count = ifelse(row_number() / n() <= 0.25, sample(1:2, n(), replace = T), 2)
  ) %>% 
  rowwise() %>% 
  mutate(
    count = nrow(data),
    value = list(sample(c(rep(1, value_count), rep(0, count - value_count)), count))
  ) %>% 
  unnest(value) %>% 
  select(-data, -value_count, -count)

   group value
   <chr> <dbl>
 1 B         0
 2 B         0
 3 B         0
 4 B         0
 5 B         1
 6 B         0
 7 A         1
 8 A         1
 9 A         0
10 D         1
11 D         0
12 D         1
13 D         0
14 D         0
15 C         1
16 C         0
17 C         0
18 C         1

看起来我被打败了,但这里有另一种方法:

library(tidyverse)

set.seed(1)

# sample can break the rules
x <- tibble(group = c(rep("A", 3),
                 rep("B", 6),
                 rep("C", 4),
                 rep("D", 5)))

# Make all 'var' =1, then set all but first of each group to 0.
xx <- x %>% group_by(group) %>%
  mutate(var = row_number()) %>%
  mutate(var = ifelse(var == 1, 1, 0))


pct_with_two <- .75 # percentage of groups with two 1's
samp_size <- floor(length(unique(xx$group)) * pct_with_two) #round down to whole number
addl_one <- sample(unique(xx$group), size = samp_size, replace = F)

xx %>% 
  mutate(var2 = case_when(
    group %in% addl_one & row_number() == 2 ~ 1, 
    TRUE ~0)) %>% 
  mutate(var = var+var2) %>% 
  select(-var2)
#> # A tibble: 18 x 2
#> # Groups:   group [4]
#>    group   var
#>    <chr> <dbl>
#>  1 A         1
#>  2 A         1
#>  3 A         0
#>  4 B         1
#>  5 B         0
#>  6 B         0
#>  7 B         0
#>  8 B         0
#>  9 B         0
#> 10 C         1
#> 11 C         1
#> 12 C         0
#> 13 C         0
#> 14 D         1
#> 15 D         1
#> 16 D         0
#> 17 D         0
#> 18 D         0

^{于2022年3月11日由 reprex package (v0.3.0)}