问题
I have a data.table object that contains multiple columns that specify unique cases. In the small example below, the variables "name", "job", and "sex" specify the unique IDs. I would like to add missing rows so that each each case has a row for each possible instance of another variable, "from" (similar to expand.grid).
library(data.table)
set.seed(1)
mydata <- data.table(name = c("john","john","john","john","mary","chris","chris","chris"),
job = c("teacher","teacher","teacher","teacher","police","lawyer","lawyer","doctor"),
sex = c("male","male","male","male","female","female","male","male"),
from = c("NYT","USAT","BG","TIME","USAT","BG","NYT","NYT"),
score = rnorm(8))
setkeyv(mydata, cols=c("name","job","sex"))
mydata[CJ(unique(name, job, sex), unique(from))]
Here's the current data.table object:
> mydata
name job sex from score
1: john teacher male NYT -0.6264538
2: john teacher male USAT 0.1836433
3: john teacher male BG -0.8356286
4: john teacher male TIME 1.5952808
5: mary police female USAT 0.3295078
6: chris lawyer female BG -0.8204684
7: chris lawyer male NYT 0.4874291
8: chris doctor male NYT 0.7383247
Here's the result I'd like:
> mydata
name job sex from score
1: john teacher male NYT -0.6264538
2: john teacher male USAT 0.1836433
3: john teacher male BG -0.8356286
4: john teacher male TIME 1.5952808
5: mary police female NYT NA
6: mary police female USAT 0.3295078
7: mary police female BG NA
8: mary police female TIME NA
9: chris lawyer female NYT -NA
10: chris lawyer female USAT -NA
11: chris lawyer female BG -0.8204684
12: chris lawyer female TIME -NA
13: chris lawyer male NYT 0.4874291
14: chris lawyer male USAT NA
15: chris lawyer male BG NA
16: chris lawyer male TIME NA
17: chris doctor male NYT 0.7383247
18: chris doctor male USAT NA
19: chris doctor male BG NA
20: chris doctor male TIME NA
Here's what I've tried:
setkeyv(mydata, cols=c("name","job","sex"))
mydata[CJ(unique(name, job, sex), unique(from))]
But I receive the following error and adding fromLast=TRUE (or FALSE) does not give me the right solution:
Error in unique.default(name, job, sex) :
'fromLast' must be TRUE or FALSE
Here are the relevant answers I've come across (but none appears to deal with multiple keyed columns): add missing rows to a data table
Efficiently inserting default missing rows in a data.table
Fastest way to add rows for missing values in a data.frame?
回答1:
A couple of possibilities are here - https://github.com/Rdatatable/data.table/pull/814
CJ.dt = function(...) {
rows = do.call(CJ, lapply(list(...), function(x) if(is.data.frame(x)) seq_len(nrow(x)) else seq_along(x)));
do.call(data.table, Map(function(x, y) x[y], list(...), rows))
}
setkey(mydata, name, job, sex, from)
mydata[CJ.dt(unique(data.table(name, job, sex)), unique(from))]
# name job sex from score
# 1: chris doctor male NYT 0.7383247
# 2: chris doctor male BG NA
# 3: chris doctor male TIME NA
# 4: chris doctor male USAT NA
# 5: chris lawyer female NYT NA
# 6: chris lawyer female BG -0.8204684
# 7: chris lawyer female TIME NA
# 8: chris lawyer female USAT NA
# 9: chris lawyer male NYT 0.4874291
#10: chris lawyer male BG NA
#11: chris lawyer male TIME NA
#12: chris lawyer male USAT NA
#13: john teacher male NYT -0.6264538
#14: john teacher male BG -0.8356286
#15: john teacher male TIME 1.5952808
#16: john teacher male USAT 0.1836433
#17: mary police female NYT NA
#18: mary police female BG NA
#19: mary police female TIME NA
#20: mary police female USAT 0.3295078
回答2:
The dev version of tidyr now has an elegant way to do this because the expand() function now supports nesting and crossing:
library(dplyr)
mydata <- data_frame(
name = c("john","john","john","john","mary","chris","chris","chris"),
job = c("teacher","teacher","teacher","teacher","police","lawyer","lawyer","doctor"),
sex = c("male","male","male","male","female","female","male","male"),
from = c("NYT","USAT","BG","TIME","USAT","BG","NYT","NYT"),
score = rnorm(8)
)
mydata %>%
expand(c(name, job, sex), from) %>%
left_join(mydata)
#> Joining by: c("name", "job", "sex", "from")
#> Source: local data frame [20 x 5]
#>
#> name job sex from score
#> 1 chris doctor male BG NA
#> 2 chris doctor male NYT 0.5448206
#> 3 chris doctor male TIME NA
#> 4 chris doctor male USAT NA
#> 5 chris lawyer female BG 1.2015173
#> 6 chris lawyer female NYT NA
#> 7 chris lawyer female TIME NA
#> 8 chris lawyer female USAT NA
#> 9 chris lawyer male BG NA
#> 10 chris lawyer male NYT -1.0930237
#> 11 chris lawyer male TIME NA
#> 12 chris lawyer male USAT NA
#> 13 john teacher male BG 1.1345461
#> 14 john teacher male NYT 1.3032946
#> 15 john teacher male TIME 2.4901830
#> 16 john teacher male USAT -1.6449096
#> 17 mary police female BG NA
#> 18 mary police female NYT NA
#> 19 mary police female TIME NA
#> 20 mary police female USAT -0.2443080
回答3:
One possibility would be to paste the columns name, job, and sex together, get the unique values, and then do CJ with the unique values of from. After that, use cSplit from library(splitstackshape) to split the pasted column back to three columns, rename those columns with setnames, and join with mydata after setting the key.
library(splitstackshape)
library(data.table)
mydata1 <- setnames(cSplit(mydata[,CJ(unique(paste(name, job, sex)),
from=unique(from))], 'V1', ' '), 2:4, c('name', 'job', 'sex'))[,
c(2:4,1), with=FALSE]
setkeyv(mydata, cols=colnames(mydata)[1:4])
mydata[mydata1]
# name job sex from score
#1: chris doctor male BG NA
#2: chris doctor male NYT 0.7383247
#3: chris doctor male TIME NA
#4: chris doctor male USAT NA
#5: chris lawyer female BG -0.8204684
#6: chris lawyer female NYT NA
#7: chris lawyer female TIME NA
#8: chris lawyer female USAT NA
#9: chris lawyer male BG NA
#10: chris lawyer male NYT 0.4874291
#11: chris lawyer male TIME NA
#12: chris lawyer male USAT NA
#13: john teacher male BG -0.8356286
#14: john teacher male NYT -0.6264538
#15: john teacher male TIME 1.5952808
#16: john teacher male USAT 0.1836433
#17: mary police female BG NA
#18: mary police female NYT NA
#19: mary police female TIME NA
#20: mary police female USAT 0.3295078
来源：https://stackoverflow.com/questions/27372027/add-missing-rows-to-data-table-according-to-multiple-keyed-columns