r - 无法使用 MonetDB 和 MonetDBLite 在 R 中组合大型调查数据帧

Question

我正在尝试连接或绑定(bind) 12 年的调查数据，每个调查数据都有数百万个观测值和一百个左右的变量，以便在考虑调查结构的同时分析年度趋势。最近发布的适用于 R 的 MonetDBLite 似乎是为这个大型数据集实现基于列的 DBMS 的绝佳机会。不过，我在合并数据集时遇到了困难。

简而言之，将 plyr::r.bind.fill 用于 DBI:dbWriteTable 会返回一个截断的表。 r.bind.fill 是理想的，因为数据集的变量每年都不同。使用 dplyr::rbind_all 似乎根本不起作用。一种不太有用的方法(但仍然可以接受)涉及删除列以使表相同并使用直接的 UNION ALL SQL 操作。但这会返回语法错误。

我是 dplyr 和 MonetDB 的新手，所以这肯定是由于我自己的无知，但我花了几天时间在网上搜索，但运气不佳，所以任何帮助将不胜感激。

以下是一个可重现的例子。

干杯

查尔斯

# install.packages("MonetDB.R", repos="http://dev.monetdb.org/Assets/R/")
# install.packages("MonetDBLite", repos="http://dev.monetdb.org/Assets/R/")
library(MonetDBLite)
library(MonetDB.R)
library(dplyr)
library(plyr)
library(zoo)


mtcars1<-mtcars[,-11] # create 2 slightly different versions of mtcars
mtcars2<-mtcars[,-10]

mtcars1<-coredata(mtcars1)[rep(seq(nrow(mtcars1)),200000),] # create large versions of those dataframes
mtcars2<-coredata(mtcars2)[rep(seq(nrow(mtcars2)),200000),]

dbdir <- tempdir() # create a temporary directory
con <- dbConnect(MonetDB.R(), embedded=dbdir) # use DBI to connect to MonetDB

dbWriteTable(con, "mtcars1", mtcars1) #  write the dataframes to column-based MonetDB tables
dbWriteTable(con, "mtcars2", mtcars2) 
dbListTables(con)

ms <- src_monetdb(embedded=dbdir) # create a dplyr::tbl version tables
mt1 <- tbl(ms, "mtcars1") 
mt2<-tbl(ms, "mtcars2")

# try plyr::rbind.fill to concatenate tables
dbWriteTable(con, "mt_1_2", rbind.fill(as.data.frame(mt1, mt2)))

# Warning message:
# Only first 6,400,000 results retrieved. Use n = -1 to retrieve all.

dbGetQuery(con, "SELECT COUNT(*) FROM mt_1_2 " )
#      L1
# 1 1e+05

dbRemoveTable(con, "mt_1_2") # remove table to re-try

# try dbFetch(res, n=-1) to retrieve all results
dbFetch(dbWriteTable(con, "mt_1_2", rbind.fill(as.data.frame(mt1, mt2))), n=-1)

# Error in (function (classes, fdef, mtable)  :
#   unable to find an inherited method for function ‘dbFetch’ for signature ‘"logical", "numeric"’
# In addition: Warning message:
# Only first 6,400,000 results retrieved. Use n = -1 to retrieve all.

dbRemoveTable(con, "mt_1_2") # remove table to re-try

dbListFields(con, "mtcars1") # remove fields to make table columns identical
dbListFields(con, "mtcars2")

dbGetQuery(con, "
ALTER TABLE mtcars1
DROP COLUMN gear
")

dbGetQuery(con, "
ALTER TABLE mtcars2
DROP COLUMN carb
")

dbGetQuery(con,
    "CREATE TABLE mt_1_2 WITH 
    Select * FROM mtcars1
    UNION ALL
    Select * FROM mtcars2")


# Error in .local(conn, statement, ...) :
#   Unable to execute statement 'CREATE TABLE mt_1_2 AS
#   Select * FROM mtcars1
#   UNION ALL
#   Select * FROM mtcars2'.
# Server says 'syntax error, unexpected SCOLON, expecting WITH in: "create table mt_1_2 as
#   select * from mtcars1
#   union all
#   select * from mtcars2"
# ' [#42000].

Answer 1

您可以坚持使用 dplyr 并使用 rbind_list

library(MonetDB.R)
library(MonetDBLite)
library(dplyr)

mtcars1 <- mtcars[, -11] # create 2 slightly different versions of mtcars
mtcars2 <- mtcars[, -10]

## Reduce size
mtcars1 <- mtcars1[rep(seq(nrow(mtcars1)), 10000), ]
mtcars2 <- mtcars2[rep(seq(nrow(mtcars2)), 10000), ]

### Check size
nrow(mtcars1)
## [1] 320000

nrow(mtcars2)
## [1] 320000

###
dbdir <- tempdir() ## create a temporary directory
con <- dbConnect(MonetDB.R(), embedded = dbdir)

###
dbWriteTable(con, name = "mtcars1", value = mtcars1)
dbWriteTable(con, name = "mtcars2", value = mtcars2)

dbListTables(con)

###
ms <- src_monetdb(embedded = dbdir) # create a dplyr::tbl version tables
mt1 <- tbl(ms, "mtcars1") 
mt2 <- tbl(ms, "mtcars2")

### You need to add `n = -1` to `as.data.frame` to retrieve all rows
dbWriteTable(con, "mt_1_2", rbind_list(as.data.frame(mt1, n = -1),
                                       as.data.frame(mt2, n = -1)))

###
dbGetQuery(con, "SELECT COUNT(*) FROM mt_1_2")
##       L1
## 1 640000