Question

我有一个格式为

的data.table

id | pet   | name  
2011-01-01 | "dog" | "a"  
2011-01-02 | "dog" | "b"  
2011-01-03 | "cat" | "c"  
2011-01-04 | "dog" | "a"  
2011-01-05 | "dog" | "some"   
2011-01-06 | "cat" | "thing"

我想执行一个聚合，连接猫出现之前出现的所有狗名，例如，

id | pet   | name   | prior  
2011-01-01 | "dog" | "a"     |  
2011-01-02 | "dog" | "b"     |  
2011-01-03 | "cat" | "c"     |  "a b"  
2011-01-04 | "dog" | "a"     |  
2011-01-05 | "dog" | "some"  |  
2011-01-06 | "cat" | "thing" | "a some"

Answer 1

尝试

 library(data.table)#v1.9.5+
 setDT(df1)[, prior:= paste(name[1:(.N-1)], collapse=' ') ,
    .(group=cumsum(c(0,diff(pet=='cat'))<0))][pet!='cat',  prior:= '']
 #            id pet  name  prior
 #1: 2011-01-01  dog     a       
 #2: 2011-01-02  dog     b       
 #3: 2011-01-03  cat     c    a b
 #4: 2011-01-04  dog     a       
 #5: 2011-01-05  dog  some       
 #6: 2011-01-06  cat thing a some

或者shift的可能解决方案（在devel版本中引入，即v1.9.5），灵感来自@David Arenburg的帖子。安装devel版本的说明是here。

 setDT(df1)[, prior := paste(name[-.N], collapse= ' '), 
    .(group=cumsum(shift(pet, fill='cat')=='cat'))][pet!='cat', prior := '']

数据

df1 <- structure(list(id = c("2011-01-01 ", "2011-01-02 ", "2011-01-03 ", 
 "2011-01-04 ", "2011-01-05 ", "2011-01-06 "), pet = c("dog", 
"dog", "cat", "dog", "dog", "cat"), name = c("a", "b", "c", "a", 
"some", "thing")), .Names = c("id", "pet", "name"), row.names = c(NA, 
-6L), class = "data.frame")

Answer 2

这是另一个选择

indx <- setDT(DT)[, list(.I[.N], paste(name[-.N], collapse = ' ')), 
                    by = list(c(0L, cumsum(pet == "cat")[-nrow(DT)]))]
DT[indx$V1, prior := indx$V2]
DT
#            id pet  name  prior
# 1: 2011-01-01 dog     a     NA
# 2: 2011-01-02 dog     b     NA
# 3: 2011-01-03 cat     c    a b
# 4: 2011-01-04 dog     a     NA
# 5: 2011-01-05 dog  some     NA
# 6: 2011-01-06 cat thing a some

Answer 3

我在我的数据集上运行了每个解决方案，并将运行时间与rbenchmark进行了比较。

我无法共享数据集，但这里有一些基本信息：

Type

比较代码，

dim(event_source_causal_parts)
[1] 311127      4

结果如下，

require(rbenchmark)
benchmark({
  event_source_causal_parts <- augmented_data_no_software[, list(PROD_ID, Source, Event_Date, Causal_Part_Number)] 
  setDT(event_source_causal_parts)[, prior := paste(Causal_Part_Number[-.N], collapse = ' '), .(group=cumsum(c(0,diff(Source == "Warranty")) < 0))][Source != 'Warranty', prior := '']
 })

benchmark({
  event_source_causal_parts <- augmented_data_no_software[, list(PROD_ID, Source, Event_Date, Causal_Part_Number)] 
  setDT(event_source_causal_parts)[, prior := paste(Causal_Part_Number[-.N], collapse = ' '), .(group=cumsum(shift(Source, fill="Warranty") == "Warranty"))][Source != 'Warranty', prior := ''] 
  })


benchmark({
  event_source_causal_parts <- augmented_data_no_software[, list(PROD_ID, Source, Event_Date, Causal_Part_Number)] 
  indx <- setDT(event_source_causal_parts)[, list(.I[.N], paste(Causal_Part_Number[-.N], collapse = " ")),
                                       by = list(c(0L, cumsum(Source == "Warranty")[-nrow(event_source_causal_parts)]))]
})

我的环境，

  replications elapsed relative user.self sys.self user.child sys.child
1          100   12.91        1     12.76     0.05         NA        NA

  replications elapsed relative user.self sys.self user.child sys.child
1          100    12.7        1     12.66     0.05         NA        NA

  replications elapsed relative user.self sys.self user.child sys.child
1          100   61.97        1     61.65        0         NA        NA

R使用英特尔MKL数学库。

根据这些结果，我认为@akrun的第二个解决方案是最快的。

我再次运行测试，但现在我使用-O3重新编译data.table并将R更新为3.2.0。结果非常不同：

R version 3.1.2 (2014-10-31)
Platform: x86_64-w64-mingw32/x64 (64-bit)

locale:
[1] LC_COLLATE=English_United States.1252  LC_CTYPE=English_United States.1252   
[3] LC_MONETARY=English_United States.1252 LC_NUMERIC=C                          
[5] LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] rbenchmark_1.0.0 stringr_0.6.2    data.table_1.9.5 vimcom_1.2-6    

loaded via a namespace (and not attached):
[1] chron_2.3-45    grid_3.1.2      lattice_0.20-30 tools_3.1.2     zoo_1.7-11

所以最好的解决方案在使用O3的新R下更快，但第二个最佳解决方案要慢得多。

使用data.table创建序列

3 个答案:

数据