R，dplyr：将出现次数作为值分配给几个 group_by() 级别的列

Question

require(plyr)
require(dplyr)

set.seed(8)
df <- data.frame(
  group = sample(c("A","B"), 10, replace=T),
  subgroup = sample(c("a", "b", "c"),10, replace=T),
  value = runif(10, -1,1)
  )
df %>% arrange(group,subgroup)

给出：

         group subgroup      value
1      A        a -0.1841505
2      A        a  0.3265360
3      A        a -0.8045035
4      A        b -0.5526222
5      B        a  0.2238653
6      B        a  0.0552373
7      B        b  0.2297515
8      B        b -0.5700525
9      B        b  0.6347312
10     B        c  0.9550054

我可以指出值是高还是低，例如：

df2<-
df %>% mutate(reg = ifelse(value > 0, "high", "low"))
df2

给出：

   group subgroup      value  reg
1      A        b -0.5526222  low
2      A        a -0.1841505  low
3      B        b  0.2297515 high
4      B        b -0.5700525  low
5      A        a  0.3265360 high
6      B        c  0.9550054 high
7      A        a -0.8045035  low
8      B        a  0.2238653 high
9      B        a  0.0552373 high
10     B        b  0.6347312 high

问题：我想获得 low.group、high.group、low.subgroup 和 high.subgroup 列，指示在该组中找到多少次高值和低值（我想到了 dplyr 的 group_by(group) 和 n()，可能与 summarise()) 和组+子组级别 (group_by(group, subgroup))。这将生成一个 6 行乘 6 列的数据框（A/B 和 a/b/c 的组合，以及列 group、subgroup、low.group、high.group,low.subgroup 和 high.subgroup)。第一列应为 (A, a, 3, 1, 2, 1)，第二列应为 (A, b, 3, 1, 1, 0) 等。我可以数数，例如通过：

df %>%
group_by(group,reg) %>%
mutate(n.group=n())

但是我如何将 n.group 分成两列 low.group 和 high.group。子组也有同样的问题。

我确信 plyr、dplyr 和 reshape2 中的函数可以进行这种组合计数和汇总，但是怎么做呢？

更新：这是我会得到的手工结果：

group   subgroup    low.group   high.group  low.subgroup    high.subgroup
A   a   3   1   2   1
A   b   3   1   1   0
A   c   3   1   0   0
B   a   1   5   0   1
B   b   1   5   1   2
B   c   1   5   0   1

Answer 1

有点冗长，但似乎符合预期：

library(dplyr)
library(tidyr)
df %>% 
  mutate(value = ifelse(value > 0, "high", "low")) %>%
  group_by(group, subgroup, value) %>%
  mutate(sub = n()) %>%
  group_by(group, value) %>%
  mutate(grp = n()) %>% 
  distinct(group, subgroup, value) %>% 
  gather(key, val, sub:grp) %>%
  unite(x, value:key, sep = ".") %>%
  spread(x, val, fill = 0)

#Source: local data frame [5 x 6]
#
#  group subgroup high.grp high.sub low.grp low.sub
#1     A        a        1        1       3       2
#2     A        b        0        0       3       1
#3     B        a        5        2       0       0
#4     B        b        5        2       1       1
#5     B        c        5        1       0       0

请注意，A-c 组合不会出现在示例数据中，因此不会出现在输出中。

Answer 2

docendo discimus 解决方案的变体 - 使用更多的 reshape2 和更少的 tidyr - 是：

library(dplyr)
library(tidyr)
library(stringr)
library(reshape2)

df %>%
 mutate(value=ifelse(value > 0, "high", "low")) %>%
 group_by(group, subgroup, value) %>%
 mutate(sub = n()) %>%
 group_by(group, value) %>%
 mutate(grp = n()) %>%
 gather(key,val,sub:grp) %>%
 mutate(val.key=str_c(value,".",key)) %>%
 distinct() %>%
 dcast(group+subgroup~val.key, value.var="val", fill=0)

R，dplyr：将出现次数作为值分配给几个 group_by() 级别的列

R, dplyr: assign number of occurence as value to column at several group_by() levels

r

count

reshape2

dplyr