用所选列的最小行数替换NA

问题描述

假设我有一个包含几种类型的列（字符，数字，ID，时间等）的数据框。我将提供一个简单的示例，如下所示：

m <- data.frame(LETTERS[1:10],LETTERS[15:24],runif(10),runif(10))
x<-c("Col1","Col2","Col3","Col4","Col5","Col6","Col7")
colnames(m)<-x
m<-as.data.frame(lapply(m,function(x) x[ sample(c(TRUE,NA),prob = c(0.75,0.25),size = length(x),replace = TRUE) ]))

> m
   Col1 Col2       Col3       Col4       Col5       Col6       Col7
1     A    O 0.09929126 0.40435352 0.15360830 0.03830400 0.80157985
2     B    P 0.50314123 0.81725456         NA 0.07054851 0.65521042
3     C <NA> 0.75798665         NA 0.04483692 0.54671014         NA
4     D    R 0.96825047 0.01875140 0.07383107         NA 0.04498563
5  <NA>    S 0.47079716 0.04181401 0.21423046         NA 0.55493444
6     F <NA>         NA         NA         NA 0.33702657 0.54989260
7     G    U 0.71947656         NA         NA 0.99142181 0.69548691
8  <NA> <NA> 0.90518907 0.20661633 0.65788523 0.05534330 0.78420756
9     I    W 0.79208514 0.63233902         NA 0.72085080         NA
10    J    X 0.39093317 0.97107464         NA 0.86417719 0.39890170

对于Col3-Col7，如果NA少于3个，我想将其替换为Col3-Col7中的最小行，否则将NA保留在那里。因此，我希望数据集看起来如下：

> m
   Col1 Col2       Col3       Col4       Col5       Col6       Col7
1     A    O 0.09929126 0.40435352 0.15360830 0.03830400 0.80157985
2     B    P 0.50314123 0.81725456 0.07054851 0.07054851 0.65521042
3     C <NA> 0.75798665 0.04483692 0.04483692 0.54671014 0.04483692
4     D    R 0.96825047 0.01875140 0.07383107 0.01875140 0.04498563
5  <NA>    S 0.47079716 0.04181401 0.21423046 0.04181401 0.55493444
6     F <NA>         NA         NA         NA 0.33702657 0.54989260
7     G    U 0.71947656 0.69548691 0.69548691 0.99142181 0.69548691
8  <NA> <NA> 0.90518907 0.20661633 0.65788523 0.05534330 0.78420756
9     I    W 0.79208514 0.63233902 0.63233902 0.72085080 0.63233902
10    J    X 0.39093317 0.97107464 0.39093317 0.86417719 0.39890170

因此，第6行以外的每一行的值均由第3-7列的每一行的最小值估算。

在我的实际数据集中，对于列18:27之间的每一行，如果NA少于4，则用列18:27的最小行替换，否则保留所有NA。

我尝试使用dplyr管道/突变/替换方法，但是我不确定如何对一列列进行操作（我的印象是您只能使用突变/替换来指定一列） 。我尝试过的一些逻辑，包括在if语句中

rowSums(is.na(.[18:27]))<4 & rowSums(is.na(.[18:27]))>0)

我已经在matrixStats包中看到了rowMins函数，但是我只是想知道是否可以使用dplyr / dataframe而不是矩阵来做到这一点。

解决方法

我建议您使用一种tidyverse方法，其中您对数据进行整形并按Col1和Col2进行分组，然后重新构建数据。在使用管道的同时，我们还可以使用mutate()创建新变量，并在创建Flag变量并计算最小值之后评估所需的条件。接下来的代码：

library(tidyverse)
#Data
m <- structure(list(Col1 = c("A","B","C","D","<NA>","F","G","I","J"),Col2 = c("O","P","R","S","U","W","X"),Col3 = c(0.09929126,0.50314123,0.75798665,0.96825047,0.47079716,NA,0.71947656,0.90518907,0.79208514,0.39093317),Col4 = c(0.40435352,0.81725456,0.0187514,0.04181401,0.20661633,0.63233902,0.97107464),Col5 = c(0.1536083,0.04483692,0.07383107,0.21423046,0.65788523,NA),Col6 = c(0.038304,0.07054851,0.54671014,0.33702657,0.99142181,0.0553433,0.7208508,0.86417719),Col7 = c(0.80157985,0.65521042,0.04498563,0.55493444,0.5498926,0.69548691,0.78420756,0.3989017)),class = "data.frame",row.names = c("1","2","3","4","5","6","7","8","9","10"))

代码：

#Reshape
m %>% pivot_longer(cols = -c(Col1,Col2)) %>%
  group_by(Col1,Col2) %>% mutate(MinVal=min(value,na.rm=T),Flag=sum(is.na(value))) %>% ungroup() %>%
  mutate(value=ifelse(is.na(value) & Flag<3,MinVal,value)) %>%
  select(-c(MinVal,Flag)) %>%
  pivot_wider(names_from = name,values_from=value)

输出：

# A tibble: 10 x 7
   Col1  Col2     Col3    Col4    Col5   Col6   Col7
   <chr> <chr>   <dbl>   <dbl>   <dbl>  <dbl>  <dbl>
 1 A     O      0.0993  0.404   0.154  0.0383 0.802 
 2 B     P      0.503   0.817   0.0705 0.0705 0.655 
 3 C     <NA>   0.758   0.0448  0.0448 0.547  0.0448
 4 D     R      0.968   0.0188  0.0738 0.0188 0.0450
 5 <NA>  S      0.471   0.0418  0.214  0.0418 0.555 
 6 F     <NA>  NA      NA      NA      0.337  0.550 
 7 G     U      0.719   0.695   0.695  0.991  0.695 
 8 <NA>  <NA>   0.905   0.207   0.658  0.0553 0.784 
 9 I     W      0.792   0.632   0.632  0.721  0.632 
10 J     X      0.391   0.971   0.391  0.864  0.399