19

使用 R 中的 data.table 包,我正在尝试使用合并方法创建两个 data.table 的笛卡尔积,就像在基础 R 中所做的那样。

在基地以下工作:

#assume this order data
orders <- data.frame(date = as.POSIXct(c('2012-08-28','2012-08-29','2012-09-01')),
                     first.name = as.character(c('John','George','Henry')),
                     last.name = as.character(c('Doe','Smith','Smith')),
                     qty = c(10,50,6))

#and these dates
dates <- data.frame(date = seq(from = as.POSIXct('2012-08-28'),
                               to = as.POSIXct('2012-09-07'), by = 'day'))

#get the unique customers
cust<-unique(orders[,c('first.name','last.name')])

#using merge from base R, get the cartesian product
merge(dates, cust, by = integer(0))

但是,相同的技术在使用 data.table 时不起作用,并且会引发此错误:

"Error in merge.data.table(dates.dt, cust.dt, by = integer(0)) : 
  A non-empty vector of column names for `by` is required."
#data.table approach
library(data.table)

orders.dt <- data.table(orders)

dates.dt <- data.table(dates)

cust.dt <- unique(orders.dt[, list(first.name, last.name)])

#try to use merge (data.table) in the same manner as base
merge(dates.dt, cust.dt, by = integer(0))
Error in merge.data.table(dates.dt, cust.dt, by = integer(0)) : 
  A non-empty vector of column names for `by` is required.

我希望结果反映所有日期的所有客户名称,就像在基础中一样,但是以以 data.table 为中心的方式进行。这可能吗?

4

3 回答 3

17

如果您首先从cust-dataframe 中的第一个和最后一个构造全名,则可以使用CJ(cross-join)。您不能使用所有三个向量,因为会有 99 个项目,而且名字会与姓氏不恰当地混合在一起。

> nrow(CJ(dates$date, cust$first.name, cust$last.name ) )
[1] 99

这将返回所需的 data.table 对象:

> CJ(dates$date,paste(cust$first.name, cust$last.name) )
            V1           V2
 1: 2012-08-28 George Smith
 2: 2012-08-28  Henry Smith
 3: 2012-08-28     John Doe
 4: 2012-08-29 George Smith
 5: 2012-08-29  Henry Smith
 6: 2012-08-29     John Doe
 7: 2012-08-30 George Smith
 8: 2012-08-30  Henry Smith
 9: 2012-08-30     John Doe
10: 2012-08-31     John Doe
11: 2012-08-31 George Smith
12: 2012-08-31  Henry Smith
13: 2012-09-01     John Doe
14: 2012-09-01 George Smith
15: 2012-09-01  Henry Smith
16: 2012-09-02 George Smith
17: 2012-09-02  Henry Smith
18: 2012-09-02     John Doe
19: 2012-09-03  Henry Smith
20: 2012-09-03     John Doe
21: 2012-09-03 George Smith
22: 2012-09-04  Henry Smith
23: 2012-09-04     John Doe
24: 2012-09-04 George Smith
25: 2012-09-05 George Smith
26: 2012-09-05  Henry Smith
27: 2012-09-05     John Doe
28: 2012-09-06 George Smith
29: 2012-09-06  Henry Smith
30: 2012-09-06     John Doe
31: 2012-09-07 George Smith
32: 2012-09-07  Henry Smith
33: 2012-09-07     John Doe
            V1           V2
于 2012-09-07T17:17:21.900 回答
7

merge.data.table(x, y)是一个包装对 的调用的便捷函数x[y],因此合并需要基于两个data.tables 中的列。(这就是该错误消息试图告诉您的内容)。

一种解决方法是向两个 data.tables 添加一个虚拟列,其唯一目的是使合并成为可能:

## Add a column "k", and append it to each data.table's vector of keyed columns.
setkeyv(cust.dt[,k:=1], c(key(cust.dt), "k"))
setkeyv(dates.dt[,k:=1], c(key(dates.dt), "k"))

## Merge and then remove the dummy column
res <- merge(dates.dt, cust.dt, by="k")
head(res[,k:=NULL])
#          date first.name last.name
# 1: 2012-08-28     George     Smith
# 2: 2012-08-28      Henry     Smith
# 3: 2012-08-28       John       Doe
# 4: 2012-08-29     George     Smith
# 5: 2012-08-29      Henry     Smith
# 6: 2012-08-29       John       Doe

## Maybe also clean up cust.dt and dates.dt    
# cust.dt[,k:=NULL]
# dates.dt[,k=NULL]
于 2012-09-07T17:01:37.193 回答
2

@JoshO'Brien 的解决方案使用merge但下面是一个类似的替代方案(AFAIK)。

?data.table::merge如果我正确理解文档,X[Y]应该比data.table::merge(X,Y)(从版本 1.8.7 开始)稍快。它参考了FAQ 2.12来解决这个问题,但是FAQ有点混乱。首先,正确的参考应该是 1.12,而不是 2.12。并且它们没有表明它们是指合并的基本版本还是 data.table 之一,或两者兼而有之。因此,这最终可能只是一个看起来更混乱的等效解决方案,或者它可能更快。

[来自 Matthew 的编辑] 谢谢:现在在 v1.8.7 中得到了改进(?merge.data.table,FAQ 1.12 并添加了新的 FAQ 2.24)

DT_orders<-data.table(date=as.POSIXct(c('2012-08-28','2012-08-29','2012-08-29','2012-09-01')),
                      first.name=as.character(c('John','John','George','Henry')),
                      last.name=as.character(c('Doe','Doe','Smith','Smith')),
                      qty=c(10,2,50,6),
                      key="first.name,last.name")

# Note that I added a second record to the orders table for John Doe, to make sure it could handle duplicate first/last name combinations.

DT_dates<-data.table(date=seq(from=as.POSIXct('2012-08-28'),
                              to=as.POSIXct('2012-09-07'),by='day'),
                     key="date")

DT_custdates<-data.table(k=1,unique(DT_dates),key="k")[unique(DT_orders)[,list(k=1,first.name,last.name)]][,k:=NULL]
于 2013-01-04T20:25:53.573 回答