如何以 10000 次重复优化 r 中的分层随机抽样

Question

我需要一个使用分层随机抽样的函数重复（10000次）从我的数据中抽样改变样本大小，计算每个样本大小的均值和标准差以及return变异系数。样本量范围为 3 到 30。到目前为止我已经写了这个但是它太慢了。我需要帮助使其运行更快因为我运行多次使用这部分代码。数据框 dt1 有大约 900 个观察值 K_level 有 6 个级别

谢谢

samp <- function(nn){
  dt1 <- as.data.table(dt1)
  dt2 <- replicate(10000, dt1[, .SD[sample(x = .N, size = nn)], by = K_level], 
          simplify =  FALSE) %>% 
  data.table::rbindlist() %>% 
  .[,.(avg=mean(Bunch_weight), Sd = sd(Bunch_weight)),.(Trt)] %>% 
  .[, cvs:= Sd/avg] 
  dt3 <-  data.table::transpose(dt2)
  colnames(dt3) <- as.character(dt3[1,])
  dt4 <- dt3 %>% .[-c(1:3),] %>% .[, sample:= paste0(nn,"mts")]
  return(dt4)
}
# use the function
zzz <- c(3:30)
dat5 <- map_df(.x = c(3:30), .f = samp)  

my data
Block Trt Matno Cycle Date.harvested Girth0 Girth100 Hands Fingers Bunch_weight    Variety K_level
  1:    B1  T2     6     1     2020-03-05      1        1     1       1            5     NFUUKA      0K
  2:    B1  T6     2     1     2020-03-05      2        2     2       1            9     KIBUZI    150K
  3:    B1  T6     3     1     2020-03-09      3        3     1       2            5     NFUUKA    150K
  4:    B1  T6    24     1     2020-02-28      4        4     2       1            9     KIBUZI    150K
  5:    B1  T6    29     1     2020-03-03      5        5     3       3           14     NFUUKA    150K
 ---                                                                                                   
780:    B3  T9    12     1     2020-05-22      4        4     4       4            8     NFUUKA      0K
781:    B3 T10    10     1     2020-05-25    145       47     5       5           17     NFUUKA      0K
782:    B3 T11    14     1     2020-05-16     27       88     4       4           13 MBWAZIRUME     75K
783:    B3 T14    25     1     2020-05-24     39      119     4       3           14    KISANSA    150K
784:    B3 T14    34     1     2020-05-17     27       28     5       3           15  NAKITEMBE    150K

expected output
 T9                T1                T6               T14               T13                T7               T15
1: 0.359418301512993 0.259396490785659 0.352112606549899 0.270098407993612  0.33255344147661 0.246297750226982 0.290376334651094
2:  0.36336940312546 0.260242995748078 0.347937570013322  0.26993786977025 0.327215546595358 0.247590005787063 0.290659581719395
                  T8                T3                T4               T18               T17               T10               T11
1: 0.203153174250691  0.31104051648633 0.308308574237779 0.352809537743834 0.380933443587759 0.345214551318585 0.265386556956891
2:  0.20127162406244 0.311140161227165 0.303006865683816 0.350513136037457  0.37965782184899 0.342121680883066  0.26389652807615
                  T5               T12               T16                T2 Sample
1: 0.424907358546752 0.262966077905422 0.292193075443918 0.366954072154349      3mts
2: 0.413114236465515 0.264733595838422 0.296869773806402  0.36574334095091      4mts

Answer 1

这是您的代码，只是稍微改了一下。我认为它会产生相同的输出，但很难说清楚，因为随机性是以不同的顺序完成的，因此重置随机种子无济于事。它应该快得多（> 10 倍）。

samp2 <- function(nn){
  dt1 <- as.data.table(dt1)
  dt2 <- dt1[, .SD[as.vector(replicate(10000, sample(.N, nn)))], by = K_level, 
    .SDcols = c('Trt', 'Bunch_weight')][, 
      .(avg=mean(Bunch_weight), Sd = sd(Bunch_weight)), by = .(Trt)]
  dt2[, cvs:= Sd/avg]
  dt3 <-  data.table::transpose(dt2)
  colnames(dt3) <- as.character(dt3[1,])
  dt4 <- dt3 %>% .[-c(1:3),] %>% .[, sample:= paste0(nn,"mts")]
  return(dt4[])
}

如何以 10000 次重复优化 r 中的分层随机抽样

How optimize stratified random sampling in r with 10000 repetitions

r

data.table

simulation

resampling