我有一个包含40个变量G1_a
,G1_b
,...到G20_a
,G20_b
(根据调查得出)的数据框。我想创建20个新变量G1
... G20
,以总结现有变量。
data <- data.frame(G1_a = c(0, 0, 0, 1, NA),
G1_b = c(0, 0, 1, 1, NA),
G2_a = c(0, 0, 0, 1, NA),
G2_b = c(0, 0, 1, 1, NA))
# Reshaping without for-loop:
data <- data %>%
mutate(G1 = case_when(
G1_a == 1 ~ "own_offer",
G1_b == 1 ~ "no_offer",
T ~ NA_character_
))
data <- data %>%
mutate(G2 = case_when(
G2_a == 1 ~ "own_offer",
G2_b == 1 ~ "no_offer",
T ~ NA_character_
))
我想在for循环中自动创建新变量,例如:
# Reshaping with for-loop:
for(i in 1:2) {
data <- data %>%
mutate(assign(paste0("G", i), case_when(
get(paste0("G", i, "_a")) == 1 ~ "own_offer",
get(paste0("G", i, "_b")) == 1 ~ "no_offer",
T ~ NA_character_
)))
}
我的问题包括两个部分:
1)是否可以将assign
与mutate
合并?我知道像mutate(df, !!varname := Petal.width * n)
(请参阅here)这样的方法可以动态分配参数名称。但是,我无法将其与要运行的数据重塑结合在一起。
2)dplyr
是否允许将paste0
与case_when
和mutate
一起使用?
答案 0 :(得分:2)
这有点棘手,但是我认为这是实现这一目标的原则。最终结果是带有所需列的数据框架,从而避免了所有get()
/ assign()
的麻烦(并且不会在工作区中堆满很多派生变量)。我们分几个步骤使用tidyr::gather()
和tidyr::spread()
更改数据框的形状(宽->长->部分宽->宽)。如果看起来不知所措,请尝试在各个中间点停止管道顺序,以查看到目前为止已取得的成就。
library(tidyr)
library(dplyr)
dds <- (dd
%>% mutate(case=seq(n())) ## need a variable to distinguish rows in original data set
%>% gather(var,val,-case) ## -> long format: {case, var={G1_a,G1_b,...}, val={0,1,NA}}
%>% separate(var,c("var","response")) ## split to "G1","G2" + "a", "b"
%>% spread(response,val) ## convert back to semi-wide: {case, var, a, b}
## now collapse rows to categorical value, as above
%>% mutate(offer=case_when(a==1 ~ "own_offer",
b==1 ~ "no_offer",
TRUE ~ NA_character_))
%>% select(-c(a,b)) ## clean up now-redundant variables
%>% spread(var,offer) ## convert back to wide format: {case, G1, G2, ...}
%>% select(-case) ## now redundant
)
G1 G2
1 <NA> <NA>
2 <NA> <NA>
3 no_offer no_offer
4 own_offer own_offer
5 <NA> <NA>