在实验设计中,为什么无法针对特定长度的处理计算 Graeco 拉丁方?

In Experimental Design, Why is Graeco Latin Square cannot be Computed for specific length of Treatments?

在实验设计中,我尝试设计一个 Graeco Latin-Square,我认为它是 Latin Square 设计的扩展版本,具有更多因素。但是,我发现它表现得很奇怪,这里是一些片段通过使用长度为 1-26

的处理 1 和 2 模拟
graeco_design_possibility <- function(test_until=20){
  library(agricolae)
  k_graeco <- seq(2,test_until,1)
  bool_possibility <- c()
  for(n in 2:test_until){
    b <- design.graeco(LETTERS[1:n], 1:n)
    if(is.null(b)){
      bool_possibility <- c(bool_possibility, FALSE)
    }else{
      bool_possibility <- c(bool_possibility, TRUE)
    }
  }
  simulation_graeco <- data.frame(number_k = k_graeco, success_run=bool_possibility)
  return(simulation_graeco)
}

当我测试这个时,模拟结果如下:(注意:k=26 之后会出现更多奇怪的错误)

g <- graeco_design_possibility(26)
g
   number_k success_run
1         2        TRUE
2         3        TRUE
3         4        TRUE
4         5        TRUE
5         6       FALSE
6         7        TRUE
7         8        TRUE
8         9        TRUE
9        10        TRUE
10       11        TRUE
11       12        TRUE
12       13        TRUE
13       14       FALSE
14       15        TRUE
15       16       FALSE
16       17        TRUE
17       18       FALSE
18       19        TRUE
19       20       FALSE
20       21        TRUE
21       22       FALSE
22       23        TRUE
23       24       FALSE
24       25        TRUE
25       26       FALSE

原来是这样,我看了文档,说这个函数只对奇数和偶数(4,8,10,12)的平方 我不太理解解释,因为模拟的结果与解释有点矛盾:6,14,16 是偶数吗?那为什么问题一直这样呢?

我去掉了开发者在design.graeco()函数中应该限制的限制,老实说我不知道​​为什么要限制特定长度的处理,这是Graeco拉丁方设计没有限制的最终结果

design_graeco_custom <- function(trt1, trt2, serie = 2, seed = 0, kinds = "Super-Duper", randomization = TRUE){
  number <- 10
  if (serie > 0) 
    number <- 10^serie
  r <- length(trt1)
  if (seed == 0) {
    genera <- runif(1)
    seed <- .Random.seed[3]
  }
  set.seed(seed, kinds)
  parameters <- list(design = "graeco", trt1 = trt1, 
                     trt2 = trt2, r = r, serie = serie, seed = seed, kinds = kinds, 
                     randomization)
  col <- rep(gl(r, 1), r)
  fila <- gl(r, r)
  fila <- as.character(fila)
  fila <- as.numeric(fila)
  plots <- fila * number + (1:r)
  C1 <- data.frame(plots, row = factor(fila), col)
  
  C2 <- C1
  a <- 1:(r * r)
  dim(a) <- c(r, r)
  for (i in 1:r) {
    for (j in 1:r) {
      k <- i + j - 1
      if (k > r) 
        k <- i + j - r - 1
      a[i, j] <- k
    }
  }
  m <- trt1
  if (randomization) 
    m <- sample(trt1, r)
  C1 <- data.frame(C1, m[a])
  m <- trt2
  if (randomization) 
    m <- sample(trt2, r)
  C2 <- data.frame(C2, m[a])
  ntr <- length(trt1)
  C1 <- data.frame(C1, B = 0)
  for (k in 1:r) {
    x <- C1[k, 4]
    i <- 1
    for (j in 1:(r^2)) {
      y <- C2[(k - 1) * r + i, 4]
      if (C1[j, 4] == x) {
        C1[j, 5] <- y
        i <- i + 1
      }
    }
  }
  
  C1[, 4] <- as.factor(C1[, 4])
  C1[, 5] <- as.factor(C1[, 5])
  names(C1)[4] <- c(paste(deparse(substitute(trt1))))
  names(C1)[5] <- c(paste(deparse(substitute(trt2))))
  outdesign <- list(parameters = parameters, 
                    sketch = matrix(paste(C1[,4], C1[,5]), 
                                    byrow = TRUE, ncol = r), book = C1)
  return(outdesign)
}

而且我还发现治疗超过 26 岁时,我决定使用额外的辅助函数来生成可能的字母:

letters_construction <- function(n=27, format_letter="upper"){
  if(n > 26 && n <= 702){
    letter_result <- NULL
    letter_comb <- NULL
    if(format_letter=="upper"){
      letter_result <- LETTERS[1:26]
      letter_comb <- expand.grid(LETTERS[1:26], LETTERS[1:26])
    }else if(format_letter=="lower"){
      letter_result <- letters[1:26]
      letter_comb <- expand.grid(letters[1:26], letters[1:26])
    }
    letter_comb$comb <- paste0(letter_comb$Var2, letter_comb$Var1)
    letter_finalcomb <- as.character(letter_comb$comb)
    n_remainder <- n-26
    letter_result <- c(letter_result, letter_finalcomb[1:n_remainder])
    return(letter_result)
  }
}

所以我可以像这样实现 Big Graeco 拉丁方设计:

b <- letters_construction(30)
design_graeco_custom(b, 1:30)