r 当组合不存在时用 null 重塑

r reshape with nulls when combination doesn't exist

meltdcast使用reshape2包有一些数据,如下。

dat <- data.frame(Name = c("Alice", "Alice", "Alice", "Alice", "Bob", "Bob", "Bob"),
              Month = c(1, 1, 1, 2, 1, 2, 2),
              Product = c("Car", "Bike", "Car", "Car", "Car", "Bike", "Bike"),
              Price = c(1000, 150, 300, 500, 2000, 200, 100))
#    Name Month Product Price
# 1 Alice     1     Car  1000
# 2 Alice     1    Bike   150
# 3 Alice     1     Car   300
# 4 Alice     2     Car   500
# 5   Bob     1     Car  2000
# 6   Bob     2    Bike   200
# 7   Bob     2    Bike   100

dat_melt <- melt(dat, id=c("Name", "Month", "Product"))
#    Name Month Product variable value
# 1 Alice     1     Car    Price  1000
# 2 Alice     1    Bike    Price   150
# 3 Alice     1     Car    Price   300
# 4 Alice     2     Car    Price   500
# 5   Bob     1     Car    Price  2000
# 6   Bob     2    Bike    Price   200
# 7   Bob     2    Bike    Price   100

dat_spread <- dcast(dat_melt, Name + Month ~ Product + variable, value.var="value", fun=sum)
#    Name Month Bike_Price Car_Price
# 1 Alice     1        150      1300
# 2 Alice     2          0       500
# 3   Bob     1          0      2000
# 4   Bob     2        300         0

我怎样才能得到这个输出,以便名称-月份-产品组合不存在的情况(例如 Alice,2,Bike)returns NULLNA 而不是 0?请注意,该解决方案适用于 Price 为 0 的情况,例如dat_spread$BikePrice[BikePrice == 0] <- NA 不可接受。

我曾尝试在 dcast 中使用匿名函数但无济于事,例如

library(dplyr)
dcast(dat_melt, Name + Month ~ Product + variable, value.var="value",
      fun.aggregate = function(x) if_else(is.na(x), NULL, sum(x)))
#  Error: `false` must be type NULL, not double 

dcast(dat_melt, Name + Month ~ Product + variable, value.var="value",
      fun.aggregate = function(x) if_else(is.na(x), 3.14, sum(x))) # then update after
# Error in vapply(indices, fun, .default) : values must be length 0,
# but FUN(X[[1]]) result is length 1

请注意,reshape2 不是必需的,因此如果您有不使用它的解决方案(例如使用 tidyverse 函数),那也很好。

您可以使用 fill 参数指定用于 dcast 中缺失组合的值:

dcast(dat_melt, Name + Month ~ Product + variable,
      value.var = "value", fun = sum, fill = NA_real_)
#>    Name Month Bike_Price Car_Price
#> 1 Alice     1        150      1300
#> 2 Alice     2         NA       500
#> 3   Bob     1         NA      2000
#> 4   Bob     2        300        NA

reprex package (v0.2.0) 创建于 2018-03-07。

(请注意,在幕后,dcast 调用 vapply,这对类型很挑剔;因此仅指定 fill = NA 是不够的,因为 typeof(NA) == "logical"并且您的值是数字:您必须明确使用 "double" NA 和 NA_real_)

作为替代方案:您还可以使用 dplyr+tidyr:

进行所有重塑
library(dplyr);
library(tidyr);
dat %>%
    group_by(Name, Month, Product) %>%
    summarise(Price = sum(Price)) %>%
    spread(Product, Price);
## A tibble: 4 x 4
## Groups:   Name, Month [4]
#  Name  Month  Bike   Car
#  <fct> <dbl> <dbl> <dbl>
#1 Alice    1.  150. 1300.
#2 Alice    2.   NA   500.
#3 Bob      1.   NA  2000.
#4 Bob      2.  300.   NA

dcast类似,spread有一个fill参数,默认为fill=NA