如何在两张桌子之间进行总结?

时间:2017-01-16 19:50:26

标签: r data.table

我有两个表需要做一个sumif。表1包含时间段,即年末年和季度(即4812等)。表2包含了367等季度的年度交易。

我需要表3来汇总一年中的所有交易,以便在年底获得累积头寸。

以下是一些示例代码,用于解释数据的外观以及输出的外观:

library(data.table)

x1 <- data.table("Name" = "LOB1", "Year" = 2000, 
                 "Quarter" = c(4, 8, 12, 16, 20, 24, 28, 32, 36))
x2 <- data.table("Name" = "LOB1", "Year" = 2000, 
                 "Quarter" = c(3, 6, 7, 9, 11, 14, 16, 20, 24), 
                 "Amount" = c(10000, 15000, -2500, 3500, -6500, 25000, 
                              11000, 9000, 7500))
x3 <- data.table("Name" = "LOB1", "Year" = 2000, 
                 "Quarter" = c(4, 8, 12, 16, 20, 24, 28, 32, 36), 
                 "Amount" = c(10000, 22500, 19500, 55500, 64500, 72000, 
                              72000, 72000, 72000))

我已尝试过mergesummarisefoverlaps,但无法弄明白。

1 个答案:

答案 0 :(得分:11)

好问题。您基本上要做的是按NameYearQuarter <= Quarter加入,同时汇总所有匹配的Amount值。这可能是使用新的非equi连接(在最新的稳定版本的data.table v-1.10.0中引入)和foverlaps(后者可能是次优的)

非Equi加入:

x2[x1, # for each value in `x1` find all the matching values in `x2`
   .(Amount = sum(Amount)), # Sum all the matching values in `Amount`
   on = .(Name, Year, Quarter <= Quarter), # join conditions
   by = .EACHI] # Do the summing per each match in `i`
#    Name Year Quarter Amount
# 1: LOB1 2000       4  10000
# 2: LOB1 2000       8  22500
# 3: LOB1 2000      12  19500
# 4: LOB1 2000      16  55500
# 5: LOB1 2000      20  64500
# 6: LOB1 2000      24  72000
# 7: LOB1 2000      28  72000
# 8: LOB1 2000      32  72000
# 9: LOB1 2000      36  72000

作为旁注,您可以在Amount(由@Frank提议)中轻松添加x1

x1[, Amount := 
  x2[x1, sum(x.Amount), on = .(Name, Year, Quarter <= Quarter), by = .EACHI]$V1
]

如果您在该表中只有三个连接列,那么这可能会很方便。

<强> foverlaps:

你提到了foverlaps,所以理论上你也可以使用这个函数来实现同样的功能。虽然我担心你很容易忘记。使用foverlaps,您需要创建一个巨大的表格,其中x2中的每个值都会多次连接到x1中的每个值,并将所有内容存储在内存中

x1[, Start := 0] # Make sure that we always join starting from Q0
x2[, Start := Quarter] # In x2 we want to join all possible rows each time 
setkey(x2, Name, Year, Start, Quarter) # set keys
## Make a huge cartesian join by overlaps and then aggregate
foverlaps(x1, x2)[, .(Amount = sum(Amount)), by = .(Name, Year, Quarter = i.Quarter)]
#    Name Year Quarter Amount
# 1: LOB1 2000       4  10000
# 2: LOB1 2000       8  22500
# 3: LOB1 2000      12  19500
# 4: LOB1 2000      16  55500
# 5: LOB1 2000      20  64500
# 6: LOB1 2000      24  72000
# 7: LOB1 2000      28  72000
# 8: LOB1 2000      32  72000
# 9: LOB1 2000      36  72000