如何将时间框架数据与非重叠部分的保留 NA 合并？答案

【问题标题】：How to merge time frame data with leaving NA for non-overlapped parts?如何将时间框架数据与非重叠部分的保留 NA 合并？
【发布时间】：2015-06-26 13:19:51
【问题描述】：

我有两个数据集（df1 和 df2），它们都由时间格式的值组成。我想做像“客观的”。通过 c("id1","id2") 合并两个数据时，我想在不重叠的时间内留下“NA”。

df1

id1    id2     click_timing 
 1      11     2015-02-03 01:00:00     
 1      11     2015-02-03 02:00:00     
 1      12     2015-02-03 03:00:00     
 1      12     2015-02-03 04:00:00     
 1      13     2015-02-03 05:10:00     
 2      34     2015-02-03 03:00:00     
 2      34     2015-02-03 04:00:00     
 2      36     2015-02-03 01:00:00
 ...

df2

id1    id2     start                         end
 1      11     2015-02-03 00:20:00     2015-02-03 00:40:00
 1      11     2015-02-03 00:50:00     2015-02-03 01:20:00
 1      13     2015-02-03 01:10:00     2015-02-03 01:40:00     
 1      13     2015-02-03 04:50:00     2015-02-03 05:30:00     
 2      34     2015-02-03 03:50:00     2015-02-03 04:10:00     
 ...

目标输出

id1    id2     click_timing                start                 end
 1      11             NA             2015-02-03 00:20:00     2015-02-03 00:40:00
 1      11     2015-02-03 01:00:00    2015-02-03 00:50:00     2015-02-03 01:20:00
 1      11     2015-02-03 02:00:00            NA                  NA
 1      12     2015-02-03 03:00:00            NA                  NA
 1      12     2015-02-03 04:00:00            NA                  NA
 1      13             NA             2015-02-03 01:10:00     2015-02-03 01:40:00     
 1      13     2015-02-03 05:10:00    2015-02-03 04:50:00     2015-02-03 05:30:00
 2      34     2015-02-03 03:00:00            NA                  NA     
 2      34     2015-02-03 04:00:00     2015-02-03 03:50:00     2015-02-03 04:10:00
 2      36     2015-02-03 01:00:00            NA                  NA
 ...

【问题讨论】：

我已经尝试通过更改 all.x=T 和 all.y=T 来使用 merge(df1, df2,by=c("id1","id2"))。我不知道它为什么不起作用，但我想离开 NA 以获得不匹配的值。

标签： r merge dataset

【解决方案1】：

难题！我认为您必须通过手动循环遍历所有 click_timing 值来计算每个 click_timing 值和每个时间段（start 和 end）之间的交集，然后使用结果索引匹配作为附加连接字段：

df1 <- data.frame(id1=c(1,1,1,1,1,2,2,2), id2=c(11,11,12,12,13,34,34,36), click_timing=as.POSIXct(c('2015-02-03 01:00:00','2015-02-03 02:00:00','2015-02-03 03:00:00','2015-02-03 04:00:00','2015-02-03 05:10:00','2015-02-03 03:00:00','2015-02-03 04:00:00','2015-02-03 01:00:00')) );
df2 <- data.frame(id1=c(1,1,1,1,2), id2=c(11,11,13,13,34), start=as.POSIXct(c('2015-02-03 00:20:00','2015-02-03 00:50:00','2015-02-03 01:10:00','2015-02-03 04:50:00','2015-02-03 03:50:00')), end=as.POSIXct(c('2015-02-03 00:40:00','2015-02-03 01:20:00','2015-02-03 01:40:00','2015-02-03 05:30:00','2015-02-03 04:10:00')) );
m <- sapply(1:nrow(df1), function(i) which(df1$id1[i]==df2$id1 & df1$id2[i] == df2$id2 & df1$click_timing[i]>=df2$start & df1$click_timing[i]<=df2$end)[1] );
merge(cbind(df1,m=m),cbind(df2,m=1:nrow(df2)),by=c('id1','id2','m'),all=T)[-3];
##    id1 id2        click_timing               start                 end
## 1    1  11                <NA> 2015-02-03 00:20:00 2015-02-03 00:40:00
## 2    1  11 2015-02-03 01:00:00 2015-02-03 00:50:00 2015-02-03 01:20:00
## 3    1  11 2015-02-03 02:00:00                <NA>                <NA>
## 4    1  12 2015-02-03 04:00:00                <NA>                <NA>
## 5    1  12 2015-02-03 03:00:00                <NA>                <NA>
## 6    1  13                <NA> 2015-02-03 01:10:00 2015-02-03 01:40:00
## 7    1  13 2015-02-03 05:10:00 2015-02-03 04:50:00 2015-02-03 05:30:00
## 8    2  34 2015-02-03 04:00:00 2015-02-03 03:50:00 2015-02-03 04:10:00
## 9    2  34 2015-02-03 03:00:00                <NA>                <NA>
## 10   2  36 2015-02-03 01:00:00                <NA>                <NA>

如果存在单个click_timing 值与多个start 和end 对相交的情况，那么此解决方案将选择较早出现的那个（即df2 中的行索引较低) 比其他匹配。

【讨论】：

【解决方案2】：

重新创建初始数据框并做一些小准备：

library(data.table)
library(lubridate)

df1<- fread("id1,id2,click_timing
1,11,2015-02-03 01:00:00
1,11,2015-02-03 02:00:00
1,12,2015-02-03 03:00:00
1,12,2015-02-03 04:00:00
1,13,2015-02-03 05:10:00
2,34,2015-02-03 03:00:00
2,34,2015-02-03 04:00:00
2,36,2015-02-03 01:00:00")

# adding a redundant click_timing2 column to use as the end range for further foverlaps() function
df1[, click_timing2:= click_timing]
df1[,c("click_timing", "click_timing2"):= list(parse_date_time(click_timing, "%Y-%m-%d %T"), parse_date_time(click_timing2, "%Y-%m-%d %T"))]


df2<- fread("id1,id2,start,end
1,11,2015-02-03 00:20:00,2015-02-03 00:40:00
1,11,2015-02-03 00:50:00,2015-02-03 01:20:00
1,13,2015-02-03 01:10:00,2015-02-03 01:40:00
1,13,2015-02-03 04:50:00,2015-02-03 05:30:00
2,34,2015-02-03 03:50:00,2015-02-03 04:10:00")

df2[,c("start","end") := list(parse_date_time(start, "%Y-%m-%d %T"), parse_date_time(end, "%Y-%m-%d %T"))]
setkey(df2, id1, id2, start, end)

解决方案：

df3<- foverlaps(df1, df2, by.x=c("id1", "id2", "click_timing", "click_timing2"), 
                          by.y = c("id1", "id2", "start", "end"), type="within")
objective_output<- merge(df3, df2, by = c("id1", "id2", "start", "end"), all = T)
# deleting redundant click_timing2 column
objective_output[,click_timing2:= NULL]
# reordering columns
setcolorder(objective_output, c(1,2,5,3,4))
#setting key using all columns and thus reordering all rows
setkey(objective_output)
objective_output
#id1 id2        click_timing               start                 end
# 1:   1  11 2015-02-03 02:00:00                <NA>                <NA>
# 2:   1  11                <NA> 2015-02-03 00:20:00 2015-02-03 00:40:00
# 3:   1  11 2015-02-03 01:00:00 2015-02-03 00:50:00 2015-02-03 01:20:00
# 4:   1  12 2015-02-03 03:00:00                <NA>                <NA>
# 5:   1  12 2015-02-03 04:00:00                <NA>                <NA>
# 6:   1  13                <NA> 2015-02-03 01:10:00 2015-02-03 01:40:00
# 7:   1  13 2015-02-03 05:10:00 2015-02-03 04:50:00 2015-02-03 05:30:00
# 8:   2  34 2015-02-03 03:00:00                <NA>                <NA>
# 9:   2  34 2015-02-03 04:00:00 2015-02-03 03:50:00 2015-02-03 04:10:00
#10:   2  36 2015-02-03 01:00:00                <NA>                <NA>

【讨论】：