r - 使用 "data.table"包合并数据帧时出错

Question

以下是可重现的示例 我遇到并坚持的情况(这是我用来评估 合并数据集 以用于我的论文研究的各种方法的测试客户端)。

testData <- "https://github.com/abnova/test/blob/master/mergeTestData.zip?raw=true"

tmpFile <- tempfile()
tmpDir <- tempdir()

download.file(testData, tmpFile, method = 'curl',
              extra = '-L', quiet = TRUE)
testFiles <- unzip(tmpFile, exdir = tmpDir)

# To enable desired merge option, uncomment corresponding line

#MERGE_OPTION <- "lapply_merge"
#MERGE_OPTION <- "lapply_merge2"
#MERGE_OPTION <- "reduce_merge"
#MERGE_OPTION <- "reduce_merge2"
#MERGE_OPTION <- "reshape"
#MERGE_OPTION <- "plyr"
#MERGE_OPTION <- "dplyr"
MERGE_OPTION <- "data.table"
#MERGE_OPTION <- "data.table2"

loadData <- function (dataFile) {

  if (file.exists(dataFile)) {
    data <- readRDS(dataFile)
  }
  else { # error() undefined - replaced for stop() for now
    stop("Data file \'", dataFile, "\' not found! Run 'make' first.")
  }
  return (data)
}

loadDataSets <- function (dataDir) {

  dataSets <- list()

  dataFiles <- dir(dataDir, pattern='\\.rds$')
  dataSets <- lapply(seq_along(dataFiles),
                     function(i) {
                       nameSplit <- strsplit(dataFiles[i], "\\.")
                       dataset <- nameSplit[[1]][1]
                       assign(dataset,
                              loadData(file.path(dataDir, dataFiles[i])))
                       return (get(dataset))
                     })
  return (dataSets)
}

# load the datasets of transformed data
dataSets <- loadDataSets(tmpDir)

if (MERGE_OPTION == "lapply_merge") { # Option 1

  flossData <- data.frame(dataSets[[1]][1])

  # merge all loaded datasets by common column ("Project ID")
  silent <- lapply(seq(2, length(dataSets)),
                   function(i) {merge(flossData, dataSets[[1]][i],
                                      by = "Project ID",
                                      all = TRUE)})
}

if (MERGE_OPTION == "lapply_merge2") { # Option 1

  pids <- which(sapply(dataSets,
                       FUN=function(x) {'Project ID' %in% names(x)}))

  flossData <- dataSets[[pids[1]]]

  for (id in pids[2:length(pids)]) {
    flossData <- merge(flossData, dataSets[[id]],
                       by='Project ID', all = TRUE)
  }
}

if (MERGE_OPTION == "reduce_merge") { # Option 2

  flossData <- Reduce(function(...) 
    merge(..., by.x = "row.names", by.y = "Project ID", all = TRUE),
    dataSets)
}

# http://r.789695.n4.nabble.com/merge-multiple-data-frames-tt4331089.html#a4333772
if (MERGE_OPTION == "reduce_merge2") { # Option 2

    mergeAll <- function(..., by = "Project ID", all = TRUE) {
    dotArgs <- list(...)
    dotNames <- lapply(dotArgs, names)
    repNames <- Reduce(intersect, dotNames)
    repNames <- repNames[repNames != by]
    for(i in seq_along(dotArgs)){
      wn <- which( (names(dotArgs[[i]]) %in% repNames) &
                     (names(dotArgs[[i]]) != by))
      names(dotArgs[[i]])[wn] <- paste(names(dotArgs[[i]])[wn],
                                       names(dotArgs)[[i]], sep = ".")
    }
    Reduce(function(x, y) merge(x, y, by = by, all = all), dotArgs)
  }

  flossData <- mergeAll(dataSets)
}

if (MERGE_OPTION == "reshape") { # Option 3

  if (!suppressMessages(require(reshape))) install.packages('reshape')
  library(reshape)
  flossData <- reshape::merge_all(dataSets)
}

if (MERGE_OPTION == "plyr") { # Option 4

  if (!suppressMessages(require(plyr))) install.packages('plyr')
  library(plyr)
  flossData <- plyr::join_all(dataSets)
}

if (MERGE_OPTION == "dplyr") { # Option 5

  if (!suppressMessages(require(dplyr))) install.packages('dplyr')
  library(dplyr)

  flossData <- dataSets[[1]][1]
  flossData <- lapply(dataSets[[1]][-1],
                      function(x) {dplyr::left_join(x, flossData)})
}

if (MERGE_OPTION == "data.table") { # Option 6

  if (!suppressMessages(require(data.table))) 
    install.packages('data.table')
  library(data.table)

  flossData <- data.table(dataSets[[1]], key="Project ID")

  for (id in 2:length(dataSets)) {
    flossData <- merge(flossData, data.table(dataSets[[id]]),
                       by='Project ID', all.x = TRUE, all.y = FALSE)
  }
}

# http://stackoverflow.com/a/17458887/2872891
if (MERGE_OPTION == "data.table2") { # Option 6

  if (!suppressMessages(require(data.table))) 
    install.packages('data.table')
  library(data.table)

  DT <- data.table(dataSets[[1]], key="Project ID")
  flossData <- lapply(dataSets[[1]][-1], function(x) DT[.(x)])
}

# Additional Transformations (see TODO above)

# convert presence of Repo URL to integer
flossData[["Repo URL"]] <- as.integer(flossData[["Repo URL"]] != "")

# convert License Restrictiveness' factor levels to integers
#flossData[["License Restrictiveness"]] <- 
#  as.integer(flossData[["License Restrictiveness"]])

# convert User Community Size from character to integer
flossData[["User Community Size"]] <- 
  as.integer(flossData[["User Community Size"]])

# remove NAs
#flossData <- flossData[complete.cases(flossData[,3]),]
rowsNA <- apply(flossData, 1, function(x) {any(is.na(x))})
flossData <- flossData[!rowsNA,]

print(str(flossData))

错误信息 如下:

Starting bmerge ...done in 0.001 secs
Starting bmerge ...done in 0.002 secs
Error in vecseq(f__, len__, if (allow.cartesian) NULL else as.integer(max(nrow(x),  : 

Join results in 121229 rows; more than 100000 = max(nrow(x),nrow(i)). Check for duplicate key values in i, each of which join to the same group in x over and over again. If that's ok, try including j and dropping by (by-without-by) so that j runs for each group to avoid the large allocation. If you are sure you wish to proceed, rerun with allow.cartesian=TRUE. Otherwise, please search for this error message in the FAQ, Wiki, Stack Overflow and datatable-help for advice.

当前问题 已启用 data.table选项，但是，因为它是同一个包，我也很感激下一个选项的建议，该选项使用 替代 data.table 语法 用于合并(尽管我觉得它太困惑了，但为了知识的完整性)。先感谢您!

Answer 1

我会以这种方式处理这个问题:

首先，有一条错误消息。它说什么？

Join results in 121229 rows; more than 100000 = max(nrow(x),nrow(i)). Check for duplicate key values in i, each of which join to the same group in x over and over again. If that's ok, try including j and dropping by (by-without-by) so that j runs for each group to avoid the large allocation. If you are sure you wish to proceed, rerun with allow.cartesian=TRUE. Otherwise, please search for this error message in the FAQ, Wiki, Stack Overflow and datatable-help for advice.

伟大的!但是我有很多我正在使用的数据集，很多包和很多功能。我必须将范围缩小到产生此错误的数据集。

一一测试:

ans1 = merge(as.data.table(dataSets[[1]]), as.data.table(dataSets[[2]]), 
                all.x=TRUE, all.y=FALSE, by="Project ID")
## works fine.

ans2 = merge(as.data.table(dataSets[[1]]), as.data.table(dataSets[[3]]), 
                all.x=TRUE, all.y=FALSE, by="Project ID")
## same error

啊哈，同样的错误。

阅读错误消息的第二行:

所以， dataSets[[3]] 似乎发生了一些事情。 .它说要检查 i 中的重复键值.让我们这样做:

dim(dataSets[[3]])
# [1] 81487     3
dim(unique(as.data.table(dataSets[[3]]), by="Project ID"))
# [1] 49999     3

所以， dataSets[[3]]具有重复的“项目 ID”值，因此对于每个重复的值，来自 dataSets[[1]] 的所有匹配行返回 - 这是第二行的第二部分解释的内容: each of which join to the same group in x over and over again .

试用中 allow.cartesian=TRUE :

我知道有重复的 key ，但仍希望继续。但是错误消息提到了我们如何继续，添加“allow.cartesian=TRUE”。

ans2 = merge(as.data.table(dataSets[[1]]), as.data.table(dataSets[[3]]), 
                all.x=TRUE, all.y=FALSE, by="Project ID", allow.cartesian=TRUE)

啊哈，现在它工作正常!那么 allow.cartesian = TRUE有什么用做？或者为什么要添加？错误消息说要在 stackoverflow 上搜索消息(在其他事情中)。

搜索 allow.cartesian=TRUE关于 SO:

搜索使我进入了这个 Why is allow.cartesian required at times when when joining data.tables with duplicate keys?问题，它解释了目的，并且在评论下还包含来自@Roland 的另一个链接: Merging data.tables uses more than 10 GB RAM这指向了所有开始它的最初问题。现在让我阅读那些帖子。

是 base::merge给出不同的结果？

现在， base::merge 是否返回不同的结果(100,000 行)？

dim(merge(dataSets[[1]], dataSets[[3]], all.x=TRUE, all.y=FALSE, by="Project ID"))
# [1] 121229      4

并不真地。它提供与使用 data.table 时相同的尺寸，但它并不关心是否有重复的键，而 data.table警告您合并结果的潜在爆炸性，并允许您做出明智的决定。