在不使用apriori()函数的情况下，用R语言实现Apriori算法

Question

我正在尝试使用R语言实现Apriori，但重要的部分不是使用函数Apriori()，我应该从头开始构建它。所以我做了代码，但我的代码有一个问题，我不能解决它。我在代码中所做的是:我实现了Fk−1 × F1方法。但我的问题是，当我试图输入二进制文件时，它抛出了一个缺少值的错误！我将缺少的值替换为0，但它仍然抛出错误！我认为将原始购物篮转换为二进制形式的问题！

下面是我的代码：

    multi_col = function(data_frame) {
  multivec = data.frame(val = rep(1,nrow(data_frame)))
  for(q in 1:ncol(data_frame)){
    multivec = multivec*data_frame[q]
  }
  return(multivec)
}

item = c("onion","potato","milk","burger","beer")
t1 = c(1,1,0,1,0)
t2 = c(0,1,1,1,0)
t3 = c(1,1,1,0,0)
t4 = c(1,1,1,0,0)
t5 = c(1,1,1,1,0)
t6 =c(1,1,1,1,1)

data_mat= rbind(t1,t2,t3,t4,t5,t6)
data_mat
colnames(data_mat)=item
data_mat # this is the example data frame i used to develop the code

data_mat = as.data.frame(data_mat)
data_mat

min.sup.thresh = 2
max.item = ncol(data_mat)
max.item

for(k in 1:max.item){

  if(ncol(data_mat)>1){

    Candi = list()
    Freq =  list()

    rm_col = numeric(0)

    C_seq = combn(c(1:ncol(data_mat)),k)

    for(i in 1:ncol(C_seq)){

      Candi[[i]] = colnames(data_mat[C_seq[,i]])

      if(sum(multi_col(data_mat[C_seq[,i]]))>=min.sup.thresh){
        Freq[[i]] = colnames(data_mat[C_seq[,i]])
      }else{
        rm_col = c(rm_col,i)
      }

    }
    data_mat=data_mat[(-rm_col)]
    print(paste("number of generated candidate itemsets","in C",k,"is",length(Candi)))
    print(Candi)
    print("****************************")
    print(paste("total number of frequent itemsets","in F",k,"is",length(Freq)))
    print(Freq)
    print("###################################################################################")


  }


}

你能给我一些关于如何做的建议吗？

Answer 1

我尝试过使用您的代码，结果显示使用k==5时，行combn(c(1:ncol(data_mat)),k)出现错误，因为data_mat只有4列。我并没有真正理解你所有的代码，但我想这是因为你在循环中修改了data_mat。我设置了一个名为tmp_data_mat的新变量，这样就不会出现错误。

另一种选择是在for-loop之外修改data_mat

还要注意，milk中缺少一个值，它是通过在您使用的sum函数中添加na.rm = TRUE来工作的。

# I create data_mat on another way
data_mat <( data.frame(onion = c(1,0,rep(1,4)), 
                       potato = rep(1,6), 
                       milk = c(NA,rep(1,5)), 
                       burger = c(1,1,0,0,1,1), 
                       beer = c(rep(0,5),1)))
data_mat

min.sup.thresh = 2
max.item = ncol(data_mat)
max.item

for(k in 1:max.item){

  if(ncol(data_mat)>1){

    Candi = list()
    Freq =  list()

    # modification here about rm_col, so it don't eat all your memory.
    rm_col = seq(ncol(data_mat))

    # here is the issue I think
    C_seq = combn(c(1:ncol(data_mat)),k)

    for(i in 1:ncol(C_seq)){

      # don't think you need a function multi_col so I put it inside
      data_frame <- data_mat[C_seq[,i]]

      Candi[[i]] = colnames(data_frame)

      multivec = data.frame(val = rep(1,nrow(data_frame)))
      for(q in 1:ncol(data_frame)){
        multivec = multivec*data_frame[q]
      }

      # the missing value error was because you missed the na.rm = TRUE in the sum function !
      if(sum(multivec, na.rm = TRUE) >= min.sup.thresh){
        Freq[[i]] = colnames(data_mat[C_seq[,i]])
      }else{

        # follow the modification of rm_col
        rm_col = rm_col[-i]
      }

    }

    # here is the BIG modification of your code it don't show error.
    tmp_data_mat=data_mat[rm_col]
    print(paste("number of generated candidate itemsets","in C",k,"is",length(Candi)))
    print(Candi)
    print("****************************")
    print(paste("total number of frequent itemsets","in F",k,"is",length(Freq)))
    print(Freq)
    print("###################################################################################")

  }

}