基于多列和阈值合并数据框答案

【问题标题】：merge dataframes based on multiple columns and thresholds基于多列和阈值合并数据框
【发布时间】：2019-11-05 16:49:35
【问题描述】：

我有两个 data.frames 和多个公共列（这里：date、city、ctry 和 (other_)number）。

我现在想将它们合并到上述列中，但可以容忍一定程度的差异：

threshold.numbers <- 3
threshold.date <- 5  # in days

如果 date 条目之间的差异是 > threshold.date（以天为单位）或 > threshold.numbers，我不希望合并这些行。同样，如果city 中的条目是city 列中另一个df 条目的子字符串，我希望合并这些行。 [如果有人有更好的想法来测试实际城市名称的相似性，我很乐意听到。]（并保留第一个 df 的 date、city 和 country 条目但 (other_)number 列和 df 中的所有其他列。

考虑以下示例：

df1 <- data.frame(date = c("2003-08-29", "1999-06-12", "2000-08-29", "1999-02-24", "2001-04-17",
                           "1999-06-30", "1999-03-16", "1999-07-16", "2001-08-29", "2002-07-30"),
                  city = c("Berlin", "Paris", "London", "Rome", "Bern",
                           "Copenhagen", "Warsaw", "Moscow", "Tunis", "Vienna"),
                  ctry = c("Germany", "France", "UK", "Italy", "Switzerland",
                           "Denmark", "Poland", "Russia", "Tunisia", "Austria"),
                  number = c(10, 20, 30, 40, 50, 60, 70, 80, 90, 100),
                  col = c("apple", "banana", "pear", "banana", "lemon", "cucumber", "apple", "peach", "cherry", "cherry"))


df2 <- data.frame(date = c("2003-08-29", "1999-06-12", "2000-08-29", "1999-02-24", "2001-04-17", # all identical to df1
                           "1999-06-29", "1999-03-14", "1999-07-17", # all 1-2 days different
                           "2000-01-29", "2002-07-01"), # all very different (> 2 weeks)
                  city = c("Berlin", "East-Paris", "near London", "Rome", # same or slight differences
                           "Zurich", # completely different
                           "Copenhagen", "Warsaw", "Moscow", "Tunis", "Vienna"), # same
                  ctry = c("Germany", "France", "UK", "Italy", "Switzerland", # all the same 
                           "Denmark", "Poland", "Russia", "Tunisia", "Austria"),
                  other_number = c(13, 17, 3100, 45, 51, 61, 780, 85, 90, 101), # slightly different to very different
                  other_col = c("yellow", "green", "blue", "red", "purple", "orange", "blue", "red", "black", "beige"))

现在，我想合并data.frames 并接收df，如果满足上述条件，则合并行。

(第一列只是为了方便：在第一个数字后面，表示原始大小写，它显示了合并的行（.）或行是否来自df1（1）或df2 (2)。

          date        city        ctry number other_col other_number    other_col2          #comment
 1.  2003-08-29      Berlin     Germany     10     apple              13        yellow      # matched on date, city, number
 2.  1999-06-12       Paris      France     20    banana              17         green      # matched on date, city similar, number - other_number == threshold.numbers
 31  2000-08-29      London          UK     30      pear            <NA>          <NA>      # not matched: number - other_number > threshold.numbers
 32  2000-08-29 near London         UK    <NA>      <NA>            3100          blue      #
 41  1999-02-24        Rome       Italy     40    banana            <NA>          <NA>      # not matched: number - other_number > threshold.numbers
 42  1999-02-24        Rome       Italy   <NA>      <NA>              45           red      #
 51  2001-04-17        Bern Switzerland     50     lemon            <NA>          <NA>      # not matched: cities different (dates okay, numbers okay)
 52  2001-04-17      Zurich Switzerland   <NA>      <NA>              51        purple      #
 6.  1999-06-30  Copenhagen     Denmark     60  cucumber              61        orange      # matched: date difference < threshold.date (cities okay, dates okay)
 71  1999-03-16      Warsaw      Poland     70     apple            <NA>          <NA>      # not matched: number - other_number > threshold.numbers (dates okay)
 72  1999-03-14      Warsaw      Poland   <NA>      <NA>             780          blue      # 
 81  1999-07-16      Moscow      Russia     80     peach            <NA>          <NA>      # not matched: number - other_number > threshold.numbers (dates okay)
 82  1999-07-17      Moscow      Russia   <NA>      <NA>              85           red      #
 91  2001-08-29       Tunis     Tunisia     90    cherry            <NA>          <NA>      # not matched: date difference < threshold.date (cities okay, dates okay)
 92  2000-01-29       Tunis     Tunisia   <NA>      <NA>              90         black      #
101  2002-07-30      Vienna     Austria    100    cherry            <NA>          <NA>      # not matched: date difference < threshold.date (cities okay, dates okay)
102  2002-07-01      Vienna     Austria   <NA>      <NA>             101         beige      #

我尝试了不同的合并方式，但无法实现阈值。

编辑对不明确的表述表示歉意 - 我想保留所有行并接收指示该行是否匹配、不匹配和来自 df1 或不匹配和来自 df2。

伪代码为：

  if there is a case where abs("date_df2" - "date_df1") <= threshold.date:
    if "ctry_df2" == "ctry_df1":
      if "city_df2" ~ "city_df1":
        if abs("number_df2" - "number_df1") <= threshold.numbers:
          merge and go to next row in df2
  else:
    add row to df1```

【问题讨论】：

这是您打印的最后一个数据帧，是您想要获得的输出吗？即最后应该有 17 行？或者只是用.标记的3？
我实际上希望保留所有行，但如果它们匹配，则带有一个指示符。抱歉，如果不清楚；我相应地编辑了问题。
所以这意味着你想要 10 行和原来一样？
我添加了伪代码以使其更清晰；这有帮助吗？
如果 data.frame 不是您唯一的选择，我强烈建议使用 data.table

标签： r dataframe

【解决方案1】：

我首先将城市名称转换为字符向量，因为（如果我理解正确的话）您希望包含包含在 df2 中的城市名称。

df1$city<-as.character(df1$city)
df2$city<-as.character(df2$city)

然后按国家合并：

df = merge(df1, df2, by = ("ctry"))

> df
          ctry     date.x     city.x number      col     date.y      city.y other_number other_col
1      Austria 2002-07-30     Vienna    100   cherry 2002-07-01      Vienna          101     beige
2      Denmark 1999-06-30 Copenhagen     60 cucumber 1999-06-29  Copenhagen           61    orange
3       France 1999-06-12      Paris     20   banana 1999-06-12  East-Paris           17     green
4      Germany 2003-08-29     Berlin     10    apple 2003-08-29      Berlin           13    yellow
5        Italy 1999-02-24       Rome     40   banana 1999-02-24        Rome           45       red
6       Poland 1999-03-16     Warsaw     70    apple 1999-03-14      Warsaw          780      blue
7       Russia 1999-07-16     Moscow     80    peach 1999-07-17      Moscow           85       red
8  Switzerland 2001-04-17       Bern     50    lemon 2001-04-17      Zurich           51    purple
9      Tunisia 2001-08-29      Tunis     90   cherry 2000-01-29       Tunis           90     black
10          UK 2000-08-29     London     30     pear 2000-08-29 near London         3100      blue

stringr 库将允许您在此处查看 city.x 是否在 city.y 内（请参阅最后一列）：

library(stringr)
df$city_keep<-str_detect(df$city.y,df$city.x) # this returns logical vector if city.x is contained in city.y (works one way)
> df
          ctry     date.x     city.x number      col     date.y      city.y other_number other_col city_keep
1      Austria 2002-07-30     Vienna    100   cherry 2002-07-01      Vienna          101     beige      TRUE
2      Denmark 1999-06-30 Copenhagen     60 cucumber 1999-06-29  Copenhagen           61    orange      TRUE
3       France 1999-06-12      Paris     20   banana 1999-06-12  East-Paris           17     green      TRUE
4      Germany 2003-08-29     Berlin     10    apple 2003-08-29      Berlin           13    yellow      TRUE
5        Italy 1999-02-24       Rome     40   banana 1999-02-24        Rome           45       red      TRUE
6       Poland 1999-03-16     Warsaw     70    apple 1999-03-14      Warsaw          780      blue      TRUE
7       Russia 1999-07-16     Moscow     80    peach 1999-07-17      Moscow           85       red      TRUE
8  Switzerland 2001-04-17       Bern     50    lemon 2001-04-17      Zurich           51    purple     FALSE
9      Tunisia 2001-08-29      Tunis     90   cherry 2000-01-29       Tunis           90     black      TRUE
10          UK 2000-08-29     London     30     pear 2000-08-29 near London         3100      blue      TRUE

然后你可以得到日期之间的天数差异：

df$dayDiff<-abs(as.POSIXlt(df$date.x)$yday - as.POSIXlt(df$date.y)$yday)

以及数字上的差异：

df$numDiff<-abs(df$number - df$other_number)

生成的数据框如下所示：

> df
          ctry     date.x     city.x number      col     date.y      city.y other_number other_col city_keep dayDiff numDiff
1      Austria 2002-07-30     Vienna    100   cherry 2002-07-01      Vienna          101     beige      TRUE      29       1
2      Denmark 1999-06-30 Copenhagen     60 cucumber 1999-06-29  Copenhagen           61    orange      TRUE       1       1
3       France 1999-06-12      Paris     20   banana 1999-06-12  East-Paris           17     green      TRUE       0       3
4      Germany 2003-08-29     Berlin     10    apple 2003-08-29      Berlin           13    yellow      TRUE       0       3
5        Italy 1999-02-24       Rome     40   banana 1999-02-24        Rome           45       red      TRUE       0       5
6       Poland 1999-03-16     Warsaw     70    apple 1999-03-14      Warsaw          780      blue      TRUE       2     710
7       Russia 1999-07-16     Moscow     80    peach 1999-07-17      Moscow           85       red      TRUE       1       5
8  Switzerland 2001-04-17       Bern     50    lemon 2001-04-17      Zurich           51    purple     FALSE       0       1
9      Tunisia 2001-08-29      Tunis     90   cherry 2000-01-29       Tunis           90     black      TRUE     212       0
10          UK 2000-08-29     London     30     pear 2000-08-29 near London         3100      blue      TRUE       0    3070

但是我们想删除在 city.y 中没有找到 city.x 的东西，其中日差大于 5 或数字差大于 3：

df<-df[df$dayDiff<=5 & df$numDiff<=3 & df$city_keep==TRUE,]

> df
     ctry     date.x     city.x number      col     date.y     city.y other_number other_col city_keep dayDiff numDiff
2 Denmark 1999-06-30 Copenhagen     60 cucumber 1999-06-29 Copenhagen           61    orange      TRUE       1       1
3  France 1999-06-12      Paris     20   banana 1999-06-12 East-Paris           17     green      TRUE       0       3
4 Germany 2003-08-29     Berlin     10    apple 2003-08-29     Berlin           13    yellow      TRUE       0       3

剩下的是上面的三行（第 1 列中包含点）。

现在我们可以从 df2 中删除我们创建的三列以及日期和城市：

> df<-subset(df, select=-c(city.y, date.y, city_keep, dayDiff, numDiff))
> df
     ctry     date.x     city.x number      col other_number other_col
2 Denmark 1999-06-30 Copenhagen     60 cucumber           61    orange
3  France 1999-06-12      Paris     20   banana           17     green
4 Germany 2003-08-29     Berlin     10    apple           13    yellow

【讨论】：

【解决方案2】：

第一步：根据“city”和“ctry”合并数据：

df = merge(df1, df2, by = c("city", "ctry"))

第 2 步：如果日期条目之间的差异 > threshold.date（以天为单位），则删除行：

date_diff = abs(as.numeric(difftime(strptime(df$date.x, format = "%Y-%m-%d"),
                                    strptime(df$date.y, format = "%Y-%m-%d"), units="days")))
index_remove = date_diff > threshold.date
df = df[-index_remove,]

第 3 步：如果数字之间的差为 > threshhold.number，则删除行：

number_diff = abs(df$number - df$other_number) 
index_remove = number_diff > threshold.numbers
df = df[-index_remove,]

在应用条件之前应该合并数据，以防行不匹配。

【讨论】：

【解决方案3】：

使用data.table 的选项（内联解释）：

library(data.table)
setDT(df1)
setDT(df2)

#dupe columns and create ranges for non-equi joins
df1[, c("n", "ln", "un", "d", "ld", "ud") := .(
    number, number - threshold.numbers, number + threshold.numbers,
    date, date - threshold.date, date + threshold.date)]
df2[, c("n", "ln", "un", "d", "ld", "ud") := .(
    other_number, other_number - threshold.numbers, other_number + threshold.numbers,
    date, date - threshold.date, date + threshold.date)]

#perform non-equi join using ctry, num, dates in both ways
res <- rbindlist(list(
    df1[df2, on=.(ctry, n>=ln, n<=un, d>=ld, d<=ud),
        .(date1=x.date, date2=i.date, city1=x.city, city2=i.city, ctry1=x.ctry, ctry2=i.ctry, number, col, other_number, other_col)],
    df2[df1, on=.(ctry, n>=ln, n<=un, d>=ld, d<=ud),
        .(date1=i.date, date2=x.date, city1=i.city, city2=x.city, ctry1=i.ctry, ctry2=x.ctry, number, col, other_number, other_col)]),
    use.names=TRUE, fill=TRUE)

#determine if cities are substrings of one and another
res[, city_match := {
    i <- mapply(grepl, city1, city2) | mapply(grepl, city2, city1)
    replace(i, is.na(i), TRUE)
}]

#just like SQL coalesce (there is a version in dev in rdatatable github)
coalesce <- function(...) Reduce(function(x, y) fifelse(!is.na(y), y, x), list(...))

#for rows that are matching or no matches to be found
ans1 <- unique(res[(city_match), .(date=coalesce(date1, date2),
    city=coalesce(city1, city2),
    ctry=coalesce(ctry1, ctry2),
    number, col, other_number, other_col)])

#for rows that are close in terms of dates and numbers but are diff cities
ans2 <- res[(!city_match), .(date=c(.BY$date1, .BY$date2),
        city=c(.BY$city1, .BY$city2),
        ctry=c(.BY$ctry1, .BY$ctry2),
        number=c(.BY$number, NA),
        col=c(.BY$col, NA),
        other_number=c(NA, .BY$other_number),
        other_col=c(NA, .BY$other_col)),
    names(res)][, seq_along(names(res)) := NULL]

#final desired output
setorder(rbindlist(list(ans1, ans2)), date, city, number, na.last=TRUE)[]

输出：

          date        city        ctry number      col other_number other_col
 1: 1999-02-24        Rome       Italy     40   banana           NA      <NA>
 2: 1999-02-24        Rome       Italy     NA     <NA>           45       red
 3: 1999-03-14      Warsaw      Poland     NA     <NA>          780      blue
 4: 1999-03-16      Warsaw      Poland     70    apple           NA      <NA>
 5: 1999-06-12  East-Paris      France     20   banana           17     green
 6: 1999-06-29  Copenhagen     Denmark     60 cucumber           61    orange
 7: 1999-07-16      Moscow      Russia     80    peach           NA      <NA>
 8: 1999-07-17      Moscow      Russia     NA     <NA>           85       red
 9: 2000-01-29       Tunis     Tunisia     NA     <NA>           90     black
10: 2000-08-29      London          UK     30     pear           NA      <NA>
11: 2000-08-29 near London          UK     NA     <NA>         3100      blue
12: 2001-04-17        Bern Switzerland     50    lemon           NA      <NA>
13: 2001-04-17      Zurich Switzerland     NA     <NA>           51    purple
14: 2001-08-29       Tunis     Tunisia     90   cherry           NA      <NA>
15: 2002-07-01      Vienna     Austria     NA     <NA>          101     beige
16: 2002-07-30      Vienna     Austria    100   cherry           NA      <NA>
17: 2003-08-29      Berlin     Germany     10    apple           13    yellow

【讨论】：

【解决方案4】：

您可以使用grepl 和ctry 简单地使用== 测试city 匹配。对于那些匹配到这里的人，您可以通过使用as.Date 转换为date 并将其与difftime 进行比较来计算日期差异。 number 的区别也是一样的。

i1 <- seq_len(nrow(df1)) #Store all rows 
i2 <- seq_len(nrow(df2))
res <- do.call(rbind, sapply(seq_len(nrow(df1)), function(i) { #Loop over all rows in df1
  t1 <- which(df1$ctry[i] == df2$ctry) #Match ctry
  t2 <- grepl(df1$city[i], df2$city[t1]) | sapply(df2$city[t1], grepl, df1$city[i]) #Match city
  t1 <- t1[t2 & abs(as.Date(df1$date[i]) - as.Date(df2$date[t1[t2]])) <=
    as.difftime(threshold.date, units = "days") & #Test for date difference
    abs(df1$number[i] - df2$other_number[t1[t2]]) <= threshold.numbers] #Test for number difference
  if(length(t1) > 0) { #Match found
    i1 <<- i1[i1!=i] #Remove row as it was found
    i2 <<- i2[i2!=t1]
    cbind(df1[i,], df2[t1,c("other_number","other_col")], match=".") 
  }
}))
rbind(res
    , cbind(df1[i1,], other_number=NA, other_col=NA, match="1")
    , cbind(df2[i2,1:3], number=NA, col=NA, other_number=df2[i2,4]
            , other_col=df2[i2,5], match="2"))
#          date        city        ctry number      col other_number other_col match
#1   2003-08-29      Berlin     Germany     10    apple           13    yellow     .
#2   1999-06-12       Paris      France     20   banana           17     green     .
#6   1999-06-30  Copenhagen     Denmark     60 cucumber           61    orange     .
#3   2000-08-29      London          UK     30     pear           NA      <NA>     1
#4   1999-02-24        Rome       Italy     40   banana           NA      <NA>     1
#5   2001-04-17        Bern Switzerland     50    lemon           NA      <NA>     1
#7   1999-03-16      Warsaw      Poland     70    apple           NA      <NA>     1
#8   1999-07-16      Moscow      Russia     80    peach           NA      <NA>     1
#9   2001-08-29       Tunis     Tunisia     90   cherry           NA      <NA>     1
#10  2002-07-30      Vienna     Austria    100   cherry           NA      <NA>     1
#31  2000-08-29 near London          UK     NA     <NA>         3100      blue     2
#41  1999-02-24        Rome       Italy     NA     <NA>           45       red     2
#51  2001-04-17      Zurich Switzerland     NA     <NA>           51    purple     2
#71  1999-03-14      Warsaw      Poland     NA     <NA>          780      blue     2
#81  1999-07-17      Moscow      Russia     NA     <NA>           85       red     2
#91  2000-01-29       Tunis     Tunisia     NA     <NA>           90     black     2
#101 2002-07-01      Vienna     Austria     NA     <NA>          101     beige     2

【讨论】：

【解决方案5】：

我们可以使用 {powerjoin} ：

library(powerjoin)

power_full_join(
  df1, 
  df2, 
  by = ~ 
      # join if one city name contains the other
    (mapply(grepl, .x$city, .y$city) | mapply(grepl, .y$city, .x$city)) &
      # and dates are close enough
      abs(difftime(.x$date, .y$date, units = "days")) <= threshold.date &
      # and numbers are close enough
      abs(.x$number - .y$other_number) <= threshold.numbers,
  conflict = dplyr::coalesce)

#>    number      col other_number other_col       date        city        ctry
#> 1      10    apple           13    yellow 2003-08-29      Berlin     Germany
#> 2      20   banana           17     green 1999-06-12       Paris      France
#> 3      60 cucumber           61    orange 1999-06-30  Copenhagen     Denmark
#> 4      30     pear           NA      <NA> 2000-08-29      London          UK
#> 5      40   banana           NA      <NA> 1999-02-24        Rome       Italy
#> 6      50    lemon           NA      <NA> 2001-04-17        Bern Switzerland
#> 7      70    apple           NA      <NA> 1999-03-16      Warsaw      Poland
#> 8      80    peach           NA      <NA> 1999-07-16      Moscow      Russia
#> 9      90   cherry           NA      <NA> 2001-08-29       Tunis     Tunisia
#> 10    100   cherry           NA      <NA> 2002-07-30      Vienna     Austria
#> 11     NA     <NA>         3100      blue 2000-08-29 near London          UK
#> 12     NA     <NA>           45       red 1999-02-24        Rome       Italy
#> 13     NA     <NA>           51    purple 2001-04-17      Zurich Switzerland
#> 14     NA     <NA>          780      blue 1999-03-14      Warsaw      Poland
#> 15     NA     <NA>           85       red 1999-07-17      Moscow      Russia
#> 16     NA     <NA>           90     black 2000-01-29       Tunis     Tunisia
#> 17     NA     <NA>          101     beige 2002-07-01      Vienna     Austria

^{由reprex package 创建于 2022-04-14 (v2.0.1)}

【讨论】：

【解决方案6】：

这是一种灵活的方法，可让您指定您选择的任何合并条件集合。

准备工作

我确保 df1 和 df2 中的所有字符串都是字符串，而不是因素（如其他几个答案中所述）。我还将日期包裹在 as.Date 中以使其成为真实日期。

指定合并条件

创建一个列表列表。主列表的每个元素都是一个标准；标准的成员是

final.col.name：我们想要在决赛桌的列名
col.name.1：df1中的列名
col.name.2：df2中的列名
exact: 布尔值；我们应该对此列进行精确匹配吗？
threshold：阈值（如果我们不进行精确匹配）
match.function：返回行是否匹配的函数（对于特殊情况，例如使用grepl 进行字符串匹配；注意此函数必须进行矢量化）

merge.criteria = list(
  list(final.col.name = "date",
       col.name.1 = "date",
       col.name.2 = "date",
       exact = F,
       threshold = 5),
  list(final.col.name = "city",
       col.name.1 = "city",
       col.name.2 = "city",
       exact = F,
       match.function = function(x, y) {
         return(mapply(grepl, x, y) |
                  mapply(grepl, y, x))
       }),
  list(final.col.name = "ctry",
       col.name.1 = "ctry",
       col.name.2 = "ctry",
       exact = T),
  list(final.col.name = "number",
       col.name.1 = "number",
       col.name.2 = "other_number",
       exact = F,
       threshold = 3)
)

合并函数

这个函数接受三个参数：我们要合并的两个数据框，以及匹配条件列表。其过程如下：

遍历匹配条件并确定哪些行对满足或不满足所有条件。（受@GKi 回答的启发，它使用行索引而不是进行完全外连接，这对于大型数据集可能会占用较少的内存。）
创建一个只包含我们想要的行的骨架数据框（匹配的情况下合并行，不匹配记录的未合并行）。
遍历原始数据框的列，并使用它们在新数据框中填充所需的列。（首先对匹配条件中出现的列执行此操作，然后对剩余的任何其他列执行此操作。）

library(dplyr)
merge.data.frames = function(df1, df2, merge.criteria) {
  # Create a data frame with all possible pairs of rows from df1 and rows from
  # df2.
  row.decisions = expand.grid(df1.row = 1:nrow(df1), df2.row = 1:nrow(df2))
  # Iterate over the criteria in merge.criteria.  For each criterion, flag row
  # pairs that don't meet the criterion.
  row.decisions$merge = T
  for(criterion in merge.criteria) {
    # If we're looking for an exact match, test for equality.
    if(criterion$exact) {
      row.decisions$merge = row.decisions$merge &
        df1[row.decisions$df1.row,criterion$col.name.1] == df2[row.decisions$df2.row,criterion$col.name.2]
    }
    # If we're doing a threshhold test, test for difference.
    else if(!is.null(criterion$threshold)) {
      row.decisions$merge = row.decisions$merge &
        abs(df1[row.decisions$df1.row,criterion$col.name.1] - df2[row.decisions$df2.row,criterion$col.name.2]) <= criterion$threshold
    }
    # If the user provided a function, use that.
    else if(!is.null(criterion$match.function)) {
      row.decisions$merge = row.decisions$merge &
        criterion$match.function(df1[row.decisions$df1.row,criterion$col.name.1],
                                 df2[row.decisions$df2.row,criterion$col.name.2])
    }
  }
  # Create the new dataframe.  Just row numbers of the source dfs to start.
  new.df = bind_rows(
    # Merged rows.
    row.decisions %>% filter(merge) %>% select(-merge),
    # Rows from df1 only.
    row.decisions %>% group_by(df1.row) %>% summarize(matches = sum(merge)) %>% filter(matches == 0) %>% select(df1.row),
    # Rows from df2 only.
    row.decisions %>% group_by(df2.row) %>% summarize(matches = sum(merge)) %>% filter(matches == 0) %>% select(df2.row)
  )
  # Iterate over the merge criteria and add columns that were used for matching
  # (from df1 if available; otherwise from df2).
  for(criterion in merge.criteria) {
    new.df[criterion$final.col.name] = coalesce(df1[new.df$df1.row,criterion$col.name.1],
                                                df2[new.df$df2.row,criterion$col.name.2])
  }
  # Now add all the columns from either data frame that weren't used for
  # matching.
  for(other.col in setdiff(colnames(df1),
                           sapply(merge.criteria, function(x) x$col.name.1))) {
    new.df[other.col] = df1[new.df$df1.row,other.col]
  }
  for(other.col in setdiff(colnames(df2),
                           sapply(merge.criteria, function(x) x$col.name.2))) {
    new.df[other.col] = df2[new.df$df2.row,other.col]
  }
  # Return the result.
  return(new.df)
}

应用函数，我们就完成了

df = merge.data.frames(df1, df2, merge.criteria)

【讨论】：