按 ID 和时间在 R 中进行简单查找
simple lookup in R by ID and Time
我有一个如下所示的数据集：
set.seed(1234)
DT<-data.table(id=c(rep(c("a","b","c","d"),5)),
year=rep(seq(from = 2010.5,to=2012.5,by = .5),each=4),
value=rnorm(20,10,1))
DT
id year value
1: a 2010.5 8.792934
2: b 2010.5 10.277429
3: c 2010.5 11.084441
4: d 2010.5 7.654302
5: a 2011.0 10.429125
6: b 2011.0 10.506056
7: c 2011.0 9.425260
8: d 2011.0 9.453368
9: a 2011.5 9.435548
10: b 2011.5 9.109962
11: c 2011.5 9.522807
12: d 2011.5 9.001614
13: a 2012.0 9.223746
14: b 2012.0 10.064459
15: c 2012.0 10.959494
16: d 2012.0 9.889715
17: a 2012.5 9.488990
18: b 2012.5 9.088805
19: c 2012.5 9.162828
20: d 2012.5 12.415835
我想为每个 ID 添加 3 个非常相似的列，value_previous_6m、value_previous_y 和 value_next_y。第 10 行应该如下所示：
id year value value_previous_6m value_previous_y value_next_y
b 2011.5 9.109962 10.50606 10.27743 9.088805
我想避免使用 plyr 函数，因为整个数据集非常大。
非常感谢，
蒂姆
编辑：我知道可以用合并函数来完成：
set.seed(1234)
DT<-data.table(id=c(rep(c("a","b","c","d"),5)),
year=rep(seq(from = 2010.5,to=2012.5,by = .5),each=4),
value=rnorm(20,10,1))
DT6mp <- copy(DT)
DT12mp <- copy(DT)
DT6mp[,year:=year-.5]
setkey(DT6mp,id,year);setkey(DT,id,year);setnames(DT6mp,"value","value6mp")
DT <- merge(DT,DT6mp,all.x=T,all.y=F,allow.cartesian=T)
DT12mp[,year:=year-1]
setkey(DT12mp,id,year);setkey(DT,id,year);setnames(DT12mp,"value","value12mp")
DT <- merge(DT,DT12mp,all.x=T,all.y=F,allow.cartesian=T)
DT
不过我觉得应该有更好的方法。
你可以使用这种方法（为了更清楚，我自愿没有聚合3列加法）：
DT[,c('value_previous_6m'):=.SD[match(year-0.5, .SD$year),value], id][
,c('value_previous_y'):= .SD[match(year-1, .SD$year),value], id][
,c('value_next_y'):=.SD[match(year+1, .SD$year),value], id][]
# id year value value_previous_6m value_previous_y value_next_y
# 1: a 2010.5 8.792934 NA NA 9.435548
# 2: b 2010.5 10.277429 NA NA 9.109962
# 3: c 2010.5 11.084441 NA NA 9.522807
# 4: d 2010.5 7.654302 NA NA 9.001614
# 5: a 2011.0 10.429125 8.792934 NA 9.223746
# 6: b 2011.0 10.506056 10.277429 NA 10.064459
# 7: c 2011.0 9.425260 11.084441 NA 10.959494
# 8: d 2011.0 9.453368 7.654302 NA 9.889715
# 9: a 2011.5 9.435548 10.429125 8.792934 9.488990
#10: b 2011.5 9.109962 10.506056 10.277429 9.088805
#11: c 2011.5 9.522807 9.425260 11.084441 9.162828
#12: d 2011.5 9.001614 9.453368 7.654302 12.415835
#13: a 2012.0 9.223746 9.435548 10.429125 NA
#14: b 2012.0 10.064459 9.109962 10.506056 NA
#15: c 2012.0 10.959494 9.522807 9.425260 NA
#16: d 2012.0 9.889715 9.001614 9.453368 NA
#17: a 2012.5 9.488990 9.223746 9.435548 NA
#18: b 2012.5 9.088805 10.064459 9.109962 NA
#19: c 2012.5 9.162828 10.959494 9.522807 NA
#20: d 2012.5 12.415835 9.889715 9.001614 NA
您可以使用 data.table 的开发版本中的 shift，它具有 lag 和 lead 选项
library(data.table)#v >= 1.9.5
#library(devtools)
#install_github("Rdatatable/data.table", build_vignettes = FALSE)
DT[,c(paste0('val_previous_', c('6m', 'y')), "val_next_y"):=c(shift(value, 1:2), shift(value, 2, type="lead")), by=id]
# id year value val_previous_6m val_previous_y val_next_y
#1: a 2010.5 8.792934 NA NA 9.435548
#2: b 2010.5 10.277429 NA NA 9.109962
#3: c 2010.5 11.084441 NA NA 9.522807
#4: d 2010.5 7.654302 NA NA 9.001614
#5: a 2011.0 10.429125 8.792934 NA 9.223746
#6: b 2011.0 10.506056 10.277429 NA 10.064459
#7: c 2011.0 9.425260 11.084441 NA 10.959494
#8: d 2011.0 9.453368 7.654302 NA 9.889715
#9: a 2011.5 9.435548 10.429125 8.792934 9.488990
#10: b 2011.5 9.109962 10.506056 10.277429 9.088805
#11: c 2011.5 9.522807 9.425260 11.084441 9.162828
#12: d 2011.5 9.001614 9.453368 7.654302 12.415835
#13: a 2012.0 9.223746 9.435548 10.429125 NA
#14: b 2012.0 10.064459 9.109962 10.506056 NA
#15: c 2012.0 10.959494 9.522807 9.425260 NA
#16: d 2012.0 9.889715 9.001614 9.453368 NA
#17: a 2012.5 9.488990 9.223746 9.435548 NA
#18: b 2012.5 9.088805 10.064459 9.109962 NA
#19: c 2012.5 9.162828 10.959494 9.522807 NA
#20: d 2012.5 12.415835 9.889715 9.001614 NA
特意加长版本以免出错。
DT[, value_previous_6m := shift(value, 1), by=id
][, value_previous_y:= shift(value, 2), by=id
][, value_next_y:= shift(value, 2, type="lead"), by=id]
这里有一个dplyr+tidyr的解决方案，速度更快。
library(dplyr)
library(tidyr)
df <- data_frame(id=c(rep(c("a","b","c","d"),5)),
year=rep(seq(from = 2010.5,to=2012.5,by = .5),each=4),
value=rnorm(20,10,1))
df %>%
arrange(id) %>%
group_by(id) %>%
mutate(value_previous_6m = lag(value),
value_previous_y = lag(value, 2),
value_next_y = lead(value, 2)) %>%
ungroup %>%
arrange(year)
运行时间：0.016 秒
Beauvel 上校的 data.table 解决方案在同一台机器上花费了 0.023 秒。
编辑：Khashaa 的 data.table 解决方案比他 dplyr 我提供的解决方案更快。用了 0.007 秒。
我有一个如下所示的数据集：
set.seed(1234)
DT<-data.table(id=c(rep(c("a","b","c","d"),5)),
year=rep(seq(from = 2010.5,to=2012.5,by = .5),each=4),
value=rnorm(20,10,1))
DT
id year value
1: a 2010.5 8.792934
2: b 2010.5 10.277429
3: c 2010.5 11.084441
4: d 2010.5 7.654302
5: a 2011.0 10.429125
6: b 2011.0 10.506056
7: c 2011.0 9.425260
8: d 2011.0 9.453368
9: a 2011.5 9.435548
10: b 2011.5 9.109962
11: c 2011.5 9.522807
12: d 2011.5 9.001614
13: a 2012.0 9.223746
14: b 2012.0 10.064459
15: c 2012.0 10.959494
16: d 2012.0 9.889715
17: a 2012.5 9.488990
18: b 2012.5 9.088805
19: c 2012.5 9.162828
20: d 2012.5 12.415835
我想为每个 ID 添加 3 个非常相似的列，value_previous_6m、value_previous_y 和 value_next_y。第 10 行应该如下所示：
id year value value_previous_6m value_previous_y value_next_y
b 2011.5 9.109962 10.50606 10.27743 9.088805
我想避免使用 plyr 函数，因为整个数据集非常大。
非常感谢， 蒂姆
编辑：我知道可以用合并函数来完成：
set.seed(1234)
DT<-data.table(id=c(rep(c("a","b","c","d"),5)),
year=rep(seq(from = 2010.5,to=2012.5,by = .5),each=4),
value=rnorm(20,10,1))
DT6mp <- copy(DT)
DT12mp <- copy(DT)
DT6mp[,year:=year-.5]
setkey(DT6mp,id,year);setkey(DT,id,year);setnames(DT6mp,"value","value6mp")
DT <- merge(DT,DT6mp,all.x=T,all.y=F,allow.cartesian=T)
DT12mp[,year:=year-1]
setkey(DT12mp,id,year);setkey(DT,id,year);setnames(DT12mp,"value","value12mp")
DT <- merge(DT,DT12mp,all.x=T,all.y=F,allow.cartesian=T)
DT
不过我觉得应该有更好的方法。
你可以使用这种方法（为了更清楚，我自愿没有聚合3列加法）：
DT[,c('value_previous_6m'):=.SD[match(year-0.5, .SD$year),value], id][
,c('value_previous_y'):= .SD[match(year-1, .SD$year),value], id][
,c('value_next_y'):=.SD[match(year+1, .SD$year),value], id][]
# id year value value_previous_6m value_previous_y value_next_y
# 1: a 2010.5 8.792934 NA NA 9.435548
# 2: b 2010.5 10.277429 NA NA 9.109962
# 3: c 2010.5 11.084441 NA NA 9.522807
# 4: d 2010.5 7.654302 NA NA 9.001614
# 5: a 2011.0 10.429125 8.792934 NA 9.223746
# 6: b 2011.0 10.506056 10.277429 NA 10.064459
# 7: c 2011.0 9.425260 11.084441 NA 10.959494
# 8: d 2011.0 9.453368 7.654302 NA 9.889715
# 9: a 2011.5 9.435548 10.429125 8.792934 9.488990
#10: b 2011.5 9.109962 10.506056 10.277429 9.088805
#11: c 2011.5 9.522807 9.425260 11.084441 9.162828
#12: d 2011.5 9.001614 9.453368 7.654302 12.415835
#13: a 2012.0 9.223746 9.435548 10.429125 NA
#14: b 2012.0 10.064459 9.109962 10.506056 NA
#15: c 2012.0 10.959494 9.522807 9.425260 NA
#16: d 2012.0 9.889715 9.001614 9.453368 NA
#17: a 2012.5 9.488990 9.223746 9.435548 NA
#18: b 2012.5 9.088805 10.064459 9.109962 NA
#19: c 2012.5 9.162828 10.959494 9.522807 NA
#20: d 2012.5 12.415835 9.889715 9.001614 NA
您可以使用 data.table 的开发版本中的 shift，它具有 lag 和 lead 选项
library(data.table)#v >= 1.9.5
#library(devtools)
#install_github("Rdatatable/data.table", build_vignettes = FALSE)
DT[,c(paste0('val_previous_', c('6m', 'y')), "val_next_y"):=c(shift(value, 1:2), shift(value, 2, type="lead")), by=id]
# id year value val_previous_6m val_previous_y val_next_y
#1: a 2010.5 8.792934 NA NA 9.435548
#2: b 2010.5 10.277429 NA NA 9.109962
#3: c 2010.5 11.084441 NA NA 9.522807
#4: d 2010.5 7.654302 NA NA 9.001614
#5: a 2011.0 10.429125 8.792934 NA 9.223746
#6: b 2011.0 10.506056 10.277429 NA 10.064459
#7: c 2011.0 9.425260 11.084441 NA 10.959494
#8: d 2011.0 9.453368 7.654302 NA 9.889715
#9: a 2011.5 9.435548 10.429125 8.792934 9.488990
#10: b 2011.5 9.109962 10.506056 10.277429 9.088805
#11: c 2011.5 9.522807 9.425260 11.084441 9.162828
#12: d 2011.5 9.001614 9.453368 7.654302 12.415835
#13: a 2012.0 9.223746 9.435548 10.429125 NA
#14: b 2012.0 10.064459 9.109962 10.506056 NA
#15: c 2012.0 10.959494 9.522807 9.425260 NA
#16: d 2012.0 9.889715 9.001614 9.453368 NA
#17: a 2012.5 9.488990 9.223746 9.435548 NA
#18: b 2012.5 9.088805 10.064459 9.109962 NA
#19: c 2012.5 9.162828 10.959494 9.522807 NA
#20: d 2012.5 12.415835 9.889715 9.001614 NA
特意加长版本以免出错。
DT[, value_previous_6m := shift(value, 1), by=id
][, value_previous_y:= shift(value, 2), by=id
][, value_next_y:= shift(value, 2, type="lead"), by=id]
这里有一个dplyr+tidyr的解决方案，速度更快。
library(dplyr)
library(tidyr)
df <- data_frame(id=c(rep(c("a","b","c","d"),5)),
year=rep(seq(from = 2010.5,to=2012.5,by = .5),each=4),
value=rnorm(20,10,1))
df %>%
arrange(id) %>%
group_by(id) %>%
mutate(value_previous_6m = lag(value),
value_previous_y = lag(value, 2),
value_next_y = lead(value, 2)) %>%
ungroup %>%
arrange(year)
运行时间：0.016 秒
Beauvel 上校的 data.table 解决方案在同一台机器上花费了 0.023 秒。
编辑：Khashaa 的 data.table 解决方案比他 dplyr 我提供的解决方案更快。用了 0.007 秒。