R 嵌套循環的更快替代方案

Question

我想執行 wilcox 測試。 我有 2 個數據框列表。 Datalist 包含 2 年期間不同觀察的數量。 Varlist 包含不同場景中的 Case 和 Control-Days。

我現在想檢查一下，每次觀察的次數與病例天數和對照天數的不同場景之間是否存在差異。 因此我使用 wilcox.test(~)。

作為 output，我想要一個 dataframe，包括大小寫和控制的 2 個值、p 值，當然還有所有列表和列名，以正確匹配結果。

我有一個具有 4 次嵌套循環的工作解決方案，但它非常慢（至少需要 10 天）。 有誰知道如何用更快的代碼解決這個問題？

set.seed(42) 
n <- 365
df1 = data.frame(Date=seq.Date(as.Date("2017-01-01"), as.Date("2018-12-31"), "day"),
                  D1 = sample(18:30, n, replace=TRUE),
                  D2 = sample(0:7, n, replace=TRUE),
                  D3 = sample(0:10, n, replace=TRUE),
                  D4 = sample(0:4, n, replace=TRUE),
                  D5 = sample(0:23, n, replace=TRUE))
set.seed(7) 
n <- 365
df2 = data.frame(Date=seq.Date(as.Date("2017-01-01"), as.Date("2018-12-31"), "day"),
                 D1 = sample(18:30, n, replace=TRUE),
                 D2 = sample(0:7, n, replace=TRUE),
                 D3 = sample(0:10, n, replace=TRUE),
                 D4 = sample(0:4, n, replace=TRUE),
                 D5 = sample(0:23, n, replace=TRUE))

set.seed(9) 
n <- 365
df3 = data.frame(Date=seq.Date(as.Date("2017-01-01"), as.Date("2018-12-31"), "day"),
                 D1 = sample(18:30, n, replace=TRUE),
                 D2 = sample(0:7, n, replace=TRUE),
                 D3 = sample(0:10, n, replace=TRUE),
                 D4 = sample(0:4, n, replace=TRUE),
                 D5 = sample(0:23, n, replace=TRUE))

Datalist = list(df1, df2, df3)

set.seed(2) 
n <- 365
Var1 = data.frame(Date=seq.Date(as.Date("2017-01-01"), as.Date("2018-12-31"), "day"),
                 V1 = sample(c("Case", "Control", NA), n, replace=TRUE),
                 V2 = sample(c(NA, "Case", "Control"), n, replace=TRUE),
                 V3 = sample(c("Control", "Case", NA), n, replace=TRUE))

set.seed(6) 
n <- 365
Var2 = data.frame(Date=seq.Date(as.Date("2017-01-01"), as.Date("2018-12-31"), "day"),
                  V1 = sample(c("Case", "Control", NA), n, replace=TRUE),
                  V2 = sample(c(NA, "Case", "Control"), n, replace=TRUE),
                  V3 = sample(c("Control", "Case", NA), n, replace=TRUE))

set.seed(23) 
n <- 365
Var3 = data.frame(Date=seq.Date(as.Date("2017-01-01"), as.Date("2018-12-31"), "day"),
                  V1 = sample(c("Case", "Control", NA), n, replace=TRUE),
                  V2 = sample(c(NA, "Case", "Control"), n, replace=TRUE),
                  V3 = sample(c("Control", "Case", NA), n, replace=TRUE))

Varlist = list(Var1, Var2, Var3) 

編輯：這是我的代碼：

Results = data.frame(matrix(ncol = 7, nrow = 0))
colnames(Results) = c("Code","ICD", "Cond", "Case", "Control", "pValue", "Ver")

for (a in 1:length(Datalist)) {
  print(names(Datalist)[a])
  for (b in 2:length(Datalist[[a]])) {
    for (c in 1:length(Varlist)) {
      for (d in 2:ncol(Varlist[[c]])){ 
        Ill = Datalist[[a]][,b]
        cutpoint = nrow(Datalist[[a]])
        Group = Varlist[[c]][,d]
        Group = Group[1:cutpoint]
        casecontrol = na.omit(data.frame(Ill, Group)) 
        wiltest = wilcox.test(casecontrol$Ill ~ casecontrol$Group)  
        stats = tapply(casecontrol$Ill,casecontrol$Group,mean) 
        
        Code = names(Datalist)[a]
        ICD = colnames(Datalist[[a]])[b]
        Cond = colnames(Varlist[[c]])[d]
        Case = round(stats[1],2)
        Control  = round(stats[2],2)
        pValue  = round(wiltest$p.value, 2)
        Ver  = names(Varlist)[c]
        
        addrow = c(Code, ICD, Case, Control, pValue, Ver)
        
        Results= rbind(Results,addrow)}}}}

Answer 1

您的代碼有兩個錯誤：

1. addrow = c(Code, ICD, Case, Control, pValue, Ver)只有 6 個元素，但Results是用 7 列創建的；
2. addrow = c(Code, ICD, Case, Control, pValue, Ver)混合字符和數字數據，將所有內容強制轉換為字符。

下面的代碼解決了這些錯誤並將執行速度提高了 2 倍。一旦糾正了上述錯誤，結果是相同的。 主要區別是保留 memory 來存儲循環之前的結果，並且只在最后創建返回 data.frame。

g <- function(Datalist, Varlist) {
  ntotal <- length(Datalist) * (length(Datalist[[1]]) - 1L) * length(Varlist) * (ncol(Varlist[[1]]) - 1L)
  Code <- character(ntotal)
  ICD <- character(ntotal)
  Cond <- character(ntotal)
  Case <- numeric(ntotal)
  Control <- numeric(ntotal)
  pValue <- numeric(ntotal)
  Ver <- character(ntotal)
  i <- 0L
  for (a in 1:length(Datalist)) {
    print(names(Datalist)[a])
    for (b in 2:length(Datalist[[a]])) {
      for (c in 1:length(Varlist)) {
        for (d in 2:ncol(Varlist[[c]])){ 
          Ill = Datalist[[a]][,b]
          cutpoint = nrow(Datalist[[a]])
          Group = Varlist[[c]][,d]
          Group = Group[1:cutpoint]
          casecontrol = na.omit(data.frame(Ill, Group)) 
          wiltest = wilcox.test(Ill ~ Group, data = casecontrol)  
          stats = tapply(casecontrol$Ill,casecontrol$Group,mean) 
         
          i <- i + 1L 
          Code[i] = names(Datalist)[a]
          ICD[i] = colnames(Datalist[[a]])[b]
          Cond[i] = colnames(Varlist[[c]])[d]
          Case[i] = round(stats[1],2)
          Control[i] = round(stats[2],2)
          pValue[i] = round(wiltest$p.value, 2)
          Ver[i] = names(Varlist)[c]
        }
      }
    }
  }
  data.frame(Code, ICD, Cond, Case, Control, pValue, Ver)
}