在日期上加入两个 data.tables，表 1 中最接近的日期严格小于第二个表中的日期

Question

从 SO (Join data.table on exact date or if not the case on the nearest less than date) 的其他地方窃取了一个虚拟示例，我希望根据第一个日期（Dt1 中的日期）严格早于第二个日期（Dt2 中的日期）加入两个表。

还关闭了 DataCombine 解决方案的“幻灯片”功能的“警告”消息，因为它可能不公平地减慢了 mtotos 解决方案的速度。

library(data.table)

Dt1 <- read.table(text="
date      x
1/26/2010,  10  
1/25/2010,  9  
1/24/2010,  9   
1/22/2010,  7    
1/19/2010,  11", header=TRUE, stringsAsFactors=FALSE)

Dt2 <- read.table(text="
date
1/26/2010   
1/23/2010   
1/20/2010", header=TRUE, stringsAsFactors=FALSE)

加入的期望结果

   date     x  
1/26/2010 - 9 # based on closest observation strictly less than date  
1/23/2010 - 7   
1/20/2010 - 11

---

两种解决方案的时间

（我为 mtoto 的解决方案保留 data.frame 格式，为 jangorecki 保留 data.table 格式）。

solution.mtoto = function(Df1, Df2)
{
  #Full outer join of two df's
  merged <- merge(Df1, Df2, by = "date", all = T, sort=T)

  # Shifting values backwards by one using 'slide' from DataCombine
  merged <- slide(merged, Var = "x", slideBy = -1, reminder = F)

  # Inner join retaining the relevant cols
  return(merge(Df2,merged)[,-2])
}

solution.jangorecki = function(Dt1, Dt2)
{
  offset.roll.join = function(Dt1, Dt2){
    Dt2[, jndate := date - 1L] # produce join column with offset
    on.exit(Dt2[, jndate := NULL]) # cleanup join col on exit
    Dt1[Dt2, .(date = i.date, x), on = c("date" = "jndate"), roll = Inf] # do rolling join
  }
  return(offset.roll.join(Dt1, Dt2))
}

res.mtoto = sapply(1:10, FUN = function(x){system.time({solution.mtoto(Df1, Df2)})})

res.jangorecki = sapply(1:10, FUN = function(x){system.time({solution.jangorecki(Dt1, Dt2)})})


> res.mtoto[c("user.self", "sys.self"),]
           [,1]  [,2]  [,3]  [,4]  [,5]  [,6]  [,7]  [,8]  [,9] [,10]
user.self 0.004 0.004 0.004 0.004 0.003 0.003 0.003 0.003 0.003 0.003
sys.self  0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

> res.jangorecki[c("user.self", "sys.self"),]
           [,1]  [,2]  [,3]  [,4]  [,5]  [,6]  [,7]  [,8]  [,9] [,10]
user.self 0.005 0.005 0.004 0.004 0.005 0.004 0.004 0.004 0.003 0.004
sys.self  0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

编辑，在 mtoto 的解决方案中不小心提到了 Dt1 而不是 Df1。现已修复。

类似的速度（在更大的数据集上可能更明显？）。我的另一个问题是我希望在第二个表中返回日期。

例如，期望的结果是：

date - x - date2
1/26/2010 - 9 - 1/25/2010
1/23/2010 - 7 - 1/22/2010
1/20/2010 - 11 - 1/19/2010

Answer 1

带有-1L 偏移量的滚动连接。

2016-04-02 更新：在当前开发版本 v1.9.7 中使用 this commit，无需创建临时列即可完成此操作。来自NEWS：

在j 中可以始终使用前缀x. 引用x 的列。这在需要 x 的列同时也是连接列时特别有用。这是一个针对 #1615 的补丁。

Dt2[, jndate := date - 1L]
Dt1[Dt2,
    .(date = i.date, orgdate = x.date, x),
    on = c("date" = "jndate"),
    roll = Inf]
#         date    orgdate  x
#1: 2010-01-26 2010-01-25  9
#2: 2010-01-23 2010-01-22  7
#3: 2010-01-20 2010-01-19 11

---

原始答案，如果您使用的是 1.9.6 或更早版本，则很有用。

library(data.table)

# data
Dt1 = fread("date      x
1/26/2010,  10  
1/25/2010,  9  
1/24/2010,  9   
1/22/2010,  7    
1/19/2010,  11")[, date := as.IDate(date, format=("%m/%d/%Y"))][]
Dt2 = fread("date
1/26/2010   
1/23/2010   
1/20/2010")[, date := as.IDate(date, format=("%m/%d/%Y"))][]

# solution
offset.roll.join = function(Dt1, Dt2){
    Dt2[, jndate := date - 1L] # produce join column with offset
    Dt1[, orgdate := date] # should not be needed after data.table#1615
    on.exit({Dt2[, jndate := NULL]; Dt1[, orgdate := NULL]}) # cleanup on exit
    Dt1[Dt2, .(date = i.date, orgdate, x), on = c("date" = "jndate"), roll = Inf] # do rolling join
}
offset.roll.join(Dt1, Dt2)
#         date    orgdate  x
#1: 2010-01-26 2010-01-25  9
#2: 2010-01-23 2010-01-22  7
#3: 2010-01-20 2010-01-19 11

Answer 2

分三步：

library(DataCombine)

#Full outer join of two df's
merged <- merge(Dt1, Dt2, by = "date", all = T)

# Shifting values backwards by one using 'slide' from DataCombine
merged <- slide(merged, Var = "x", slideBy = -1)

# Inner join retaining the relevant cols
merge(Dt2,merged)[,-2]
#       date x-1
#1 1/20/2010  11
#2 1/23/2010   7
#3 1/26/2010   9