來自單個組的 dplyr sample_n

Question

我有一些數據，其中觀察次數的摘要如下所示：

# A tibble: 14 x 3
# Groups:   status [2]
   status  year     n
    <dbl> <dbl> <int>
 1      0  2010  4593
 2      0  2011 10990
 3      0  2012 27711
 4      0  2013 99989
 5      0  2014 95407
 6      0  2015 89010
 7      0  2016 72289
 8      1  2010   584
 9      1  2011   785
10      1  2012   640
11      1  2013   667
12      1  2014   377
13      1  2015   460
14      1  2016   104

其中一個組的等級明顯高於另一組的等級。 如何在不對 1 類做任何事情的情況下隨機抽樣 0 類。也就是說，我想保留所有 1 類觀測值，並通過 4593（這是該年的最小觀測數）對 0 類觀測值進行隨機采樣)

使用group_by(status, year)然后使用sample_n()不起作用，因為 4593 值大於 1 類組中的值。

我的數據的一些隨機樣本：

    structure(list(status = c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 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, 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, 1, 1, 1, 1, 1, 1, 1, 1), 
    year = c(2013, 2014, 2012, 2013, 2016, 2013, 2015, 2014, 
    2013, 2016, 2015, 2016, 2011, 2014, 2016, 2012, 2013, 2012, 
    2014, 2014, 2012, 2012, 2012, 2016, 2016, 2012, 2016, 2015, 
    2013, 2014, 2015, 2013, 2015, 2015, 2014, 2015, 2011, 2014, 
    2013, 2012, 2011, 2016, 2015, 2015, 2015, 2014, 2012, 2013, 
    2015, 2012, 2015, 2016, 2015, 2013, 2014, 2014, 2014, 2013, 
    2013, 2016, 2016, 2013, 2015, 2012, 2014, 2014, 2013, 2015, 
    2014, 2016, 2016, 2014, 2012, 2016, 2013, 2010, 2011, 2014, 
    2016, 2013, 2016, 2014, 2014, 2013, 2013, 2013, 2016, 2016, 
    2012, 2014, 2013, 2015, 2016, 2013, 2013, 2015, 2013, 2014, 
    2013, 2015, 2013, 2013, 2011, 2014, 2016, 2013, 2010, 2012, 
    2014, 2012, 2011, 2011, 2013, 2015, 2014, 2010, 2010, 2013, 
    2010, 2014, 2011, 2011, 2014, 2013, 2014, 2015, 2015, 2013, 
    2014, 2013, 2011, 2013, 2014, 2013, 2011, 2013, 2012, 2015, 
    2012, 2012, 2012, 2010, 2013, 2013, 2011, 2011, 2011, 2012, 
    2016, 2013, 2011, 2011, 2012, 2012, 2014, 2010, 2013, 2014, 
    2011, 2012, 2010, 2012, 2012, 2011, 2015, 2011, 2011, 2013, 
    2015, 2010, 2015, 2011, 2015, 2015, 2012, 2012, 2013, 2012, 
    2014, 2014, 2012, 2012, 2014, 2010, 2011, 2013, 2014, 2012, 
    2013, 2016, 2014, 2012, 2012, 2013, 2010, 2012, 2013, 2014, 
    2014, 2011)), groups = structure(list(status = c(0, 1), .rows = structure(list(
    1:100, 101:200), ptype = integer(0), class = c("vctrs_list_of", 
"vctrs_vctr"))), row.names = c(NA, -2L), class = c("tbl_df", 
"tbl", "data.frame"), .drop = TRUE), row.names = c(NA, -200L), class = c("grouped_df", 
"tbl_df", "tbl", "data.frame"))

Answer 1

我認為這會奏效。 dat是您的示例數據框。 下面的代碼通過status分割數據幀，然后使用imap來評估是否需要采樣。 如果列表元素的名稱為"0" ，則進行采樣。 您可以將size = 1更改為實際數據框的最小數量。

library(dplyr)
library(purrr)

dat2 <- dat %>%
  split(f = .$status) %>%
  imap(function(x, y){
    if (y %in% "0"){
      x <- x %>% 
        group_by(status, year) %>%
        sample_n(size = 1) 
    }
    return(x)
  }) %>%
  bind_rows()