顶级类别的一种热编码，NA，其余归入 R 中的“其他”

Question

我想对我的变量进行一次热编码，只针对顶级类别以及 NA 和“其他”。

所以在这个简化的例子中，热编码 b where freq > 1 and NA:

id <- c(1, 2, 3, 4, 5, 6)
b <- c(NA, "A", "C", "A", "B", "C")
c <- c(2, 3, 6, NA, 4, 7)
df <- data.frame(id, b, c)

  id    b  c
1  1 <NA>  2
2  2    A  3
3  3    C  6
4  4    A NA
5  5    B  4
6  6    C  7

table <- as.data.frame(table(df$b))

  Var1 Freq
1    A    2
2    B    1
3    C    2

table_top <- table[table$Freq > 1,]

  Var1 Freq
1    A    2
3    C    2

现在，我想要这样的东西

  id  b_NA  c b_A b_C b_Others
    1    1  2   0   0        0
    2    0  3   1   0        0
    3    0  6   0   1        0
    4    0 NA   1   0        0
    5    0  4   0   0        1
    6    0  7   0   1        0

我已经尝试使用子集 df

table_top <- as.vector(table_top$Var1)
table_only_top <- subset(df, b %in% table_top)
table_only_top

  a b  c
2 1 A  3
3 2 C  6
4 2 A NA
6 3 C  7

但是，现在我不知道如何获得输出。在我的真实数据中，我有比这里更多的类别，因此不能选择使用输出中的名称。我的真实输出中的其他类别也存在许多类别。

高度赞赏任何提示:)

Answer 1

data.table 和 mltools 快速而性感：

> one_hot(dt, naCols = TRUE, sparsifyNAs = TRUE)

   id cat_NA cat_A cat_C cat_Others freq
1:  1      1     0     0          0    2
2:  2      0     1     0          0    3
3:  3      0     0     1          0    6
4:  4      0     1     0          0   NA
5:  5      0     0     0          1    4
6:  6      0     0     1          0    7

代码

加载库

library(dplyr)
library(data.table)
library(mltools)

转换数据

# Kick out all with freq == 1 and below
df <- df %>%
    # Group by variables that will be onehotted
    group_by(cat) %>%
    # Add a count per group item column 
    mutate(count = n()) %>%
    # Ungroup for next steps
    ungroup() %>%
    # Change all that have a count of 1 or below to "Others".
    # If cat was a factor, we would get numeric results at this step.
    mutate(cat = ifelse(!is.na(cat) & count <= 1, "Others", cat),
    # Only now we turn it into a factor for the one_hot function 
                        cat = as.factor(cat)) %>%
    # Drop the count column
    select(id, cat, freq)

# Turn into data.table
dt <- as.data.table(df)

检查中间结果

> dt
       id    cat freq
1:  1   <NA>    2
2:  2      A    3
3:  3      C    6
4:  4      A   NA
5:  5 Others    4
6:  6      C    7

数据

id <- c(1, 2, 3, 4, 5, 6)
cat <- c(NA, "A", "C", "A", "B", "C")
freq <- c(2, 3, 6, NA, 4, 7)
# It is important to have no other factor variables other
# than the variable(s) you one want to one hot. For that reason
# the automatic factoring is turned off.
df <- data.frame(id, cat, freq, 
                 stringsAsFactors = FALSE)          

> df
  id  cat freq
1  1 <NA>    2
2  2    A    3
3  3    C    6
4  4    A   NA
5  5    B    4
6  6    C    7

Answer 2

绝对不是一个优雅的解决方案，但它应该可以工作：

library(tideverse)
library(reshape2)

df %>% 
  gather(var, val, -id) %>%
  add_count(var, val) %>% 
  mutate(res = ifelse(var == "b" & n > 1, 1, 0),
         val = paste("b_", val, sep = "")) %>% 
  filter(var == "b" & n != 1) %>% 
  dcast(id ~ val, value.var = "res") %>% 
  full_join(df, by = c("id" = "id")) %>%
  mutate(b_NA = ifelse(is.na(b), 1, 0)) %>%
  mutate_at(vars(contains("b_")), funs(replace(., is.na(.), 0))) %>%
  mutate(b_OTHERS = ifelse(rowSums(.[grep("b_", names(.))]) != 0, 0, 1))

  id b_A b_C    b  c b_NA b_OTHERS
1  2   1   0    A  3    0        0
2  3   0   1    C  6    0        0
3  4   1   0    A NA    0        0
4  6   0   1    C  7    0        0
5  1   0   0 <NA>  2    1        0
6  5   0   0    B  4    0        1

Answer 3

您可以根据您的不同标准cbinddata.frames。

# simple conditions -------------------------------------------------------
df <-  df_orig[,-1]
df_na <- is.na(df)
colnames(df_na) <- paste0(colnames(df),"_NA")
df_A <- df=="A"
colnames(df_A) <- paste0(colnames(df),"_A")
df_C <- df=="C"
colnames(df_C) <- paste0(colnames(df),"_C")

# for counts you can use sapply with one loop -----------------------------
df_counts <- df
for(j in 1:ncol(df)) {
  counts <- sapply(1:nrow(df), function(x) sum(df[x,j]==df[,j], na.rm=T) )
 df_counts[,j] <- counts
}

df_counts <- df
# or avoid explicit loops altogether --------------------------------------
df_counts2 <- sapply(1:ncol(df), function(y) sapply(1:nrow(df), function(x) sum(df[x,y]==df[,y], na.rm=T) ) )
colnames(df_counts2 ) <- paste0(colnames(df),"_counts")

# cbind df's  -------------------------------------------------------------
df_full <- cbind(df_orig, df_na, df_A, df_C, df_counts2)
# check if frequency greater then 1 or NA ---------------------------------
df_full$result <- df_full[,10:11] >=2 | df_full[,4:5]
df_full

更难的部分是我想计算频率，这里我包括了两种方法。结果是：

  id    b  c  b_NA  c_NA   b_A   c_A   b_C   c_C b_counts c_counts result.b_NA result.c_NA
1  1 <NA>  2 FALSE FALSE FALSE FALSE FALSE FALSE        1        1       FALSE       FALSE
2  2    A  3 FALSE FALSE  TRUE FALSE FALSE FALSE        2        1        TRUE       FALSE
3  3    C  6 FALSE FALSE FALSE FALSE  TRUE FALSE        2        1        TRUE       FALSE
4  4    A NA FALSE  TRUE  TRUE    NA FALSE    NA        2        0        TRUE        TRUE
5  5    B  4 FALSE FALSE FALSE FALSE FALSE FALSE        1        1       FALSE       FALSE
6  6    C  7 FALSE FALSE FALSE FALSE  TRUE FALSE        2        1        TRUE       FALSE

您可以根据自己的条件修改列。希望有帮助