Question

我有一些代码可以识别数据框中的异常值，然后删除或封顶它们。我正在尝试使用apply（）函数（或者可能是另一种方法）来加速删除过程。

示例数据

https://github.com/crossfitAL/so_ex_data/blob/master/subset
# this is the contents of a csv file, you will need to load it into your R session.

# set up an example decision-matrix
# rm.mat is a {length(cols) x 4} matrix -- in this example 8 x 4
# rm.mat[,1:2] - identify the values for min/max outliers, respectively.
# rm.mat[,3:4] - identify if you wish to remove min/max outliers, respectively.
cols <- c(1, 6:12) # specify the columns you wish to examine
rm.mat <- matrix(nrow = length(cols), ncol= 4, 
                dimnames= list(names(fico2[cols]), 
                c("out.min", "out.max","rm outliers?", "rm outliers?")))

# add example decision criteria
rm.mat[, 1] <- apply(fico2[, cols], 2, quantile, probs= .05)
rm.mat[, 2] <- apply(fico2[, cols], 2, quantile, probs= .95)
rm.mat[, 3] <- replicate(4, c(0,1))
rm.mat[, 4] <- replicate(4, c(1,0))

以下是我目前的子集代码：

df2 <- fico2 # create a copy of the data frame
cnt <- 1     # add a count variable
for (i in cols) { 
# for each column of interest in the data frame. Determine if there are min/max 
# outliers  that you wish to remove, remove them.        
  if (rm.mat[cnt, 3] == 1 & rm.mat[cnt, 4] == 1) {
    # subset / remove min and max outliers
    df2 <- df2[df2[, i] >= rm.mat[cnt, 1] & df2[, i] <= rm.mat[cnt, 2], ]  
  } else if (rm.mat[cnt, 3] == 1 & rm.mat[cnt, 4] == 0) {
    # subset / remove min outliers
    df2 <- df2[df2[, i] >= rm.mat[cnt, 1], ]
  } else if (rm.mat[cnt, 3] == 0 & rm.mat[cnt, 4] == 1) {
    # subset / remove max outliers
    df2 <- df2[df2[, i] <= rm.mat[cnt, 2], ]
  }
  cnt <- cnt + 1
}

建议的解决方案：我想我应该能够通过应用类型函数来实现这一点，删除for循环/矢量化可以加快代码速度。我遇到的问题是我正在尝试应用函数 if-and-only-if 决策矩阵表明我应该这样做。 IE-使用逻辑向量rm.mat[,3] or rm.mat[,4]来确定是否应将子集"["应用于数据帧df2。

您将获得任何帮助将不胜感激！另外，如果示例数据/代码足够，请告诉我。

Answer 1

这是一个解决方案。只是为了澄清你的代码。希望其他人可以用它来提供更好的解决方案。

因此，如果理解，您有一个决策矩阵，如下所示：

rm.mat
                                      c1 c2 c3 c4
amount.funded.by.investors     27925.000 NA  0  1
monthly.income                 11666.670 NA  1  0
open.credit.lines                 18.000 NA  0  1
revolving.credit.balance       40788.750 NA  1  0
inquiries.in.the.last.6.months     3.000 NA  0  1
debt.to.inc                       28.299 NA  1  0
int.rate                          20.490 NA  0  1
fico.num                         775.000 NA  1  0

您尝试根据此矩阵的值

过滤大矩阵

colnames(rm.mat) <- paste('c',1:4,sep='')    
rm.mat <- as.data.frame(rm.mat)
apply(rm.mat,1,function(y){
     h <- paste(y['c3'],y['c4'],sep='')
     switch(h,
            '11'= apply(df2,2, function(x)
                               df2[x >= y['c1'] &  x <= y['c2'],]),  ## we never have this!!
            '10'= apply(df2,2, function(x)
                               df2[x >= y['c1'] , ]),   ## here we apply by columns!
            '01'= apply(df2,2,function(x) 
                               df2[x <= y['c2'], ]))   ## c2 is NA!! so !!!
 }
)

在许多列上快速进行子集化

1 个答案: