RSQLite中的因素

Question

我一直無法找到RSQLite如何處理因素的文檔。從一個快速測試（見下文），它看起來像他們轉換爲字符。

問題1：有什麼方法可以將它們作爲因子保存下來嗎？我可以想到一些方法（主要涉及一個單獨的表或.Rdata文件存儲因子水平），但它似乎應該有一個標準，因此更可維護的方式來做到這一點。

問題2：如果不是RSQLite，比其他數據庫或類似數據庫的包？這裏我的用例很簡單：在處理每一個大型數據庫時添加一堆大的（2-5mm行X 550列）data.frames，以便構建一個巨大的數據庫，然後只能從該數據庫中選擇想要的數據行到data.table並繼續工作。

library(RSQLite) 
# Create 
db <- dbConnect(SQLite(), dbname="~/temp/test.sqlite") 
# Write test 
set.seed(1) 
testDat <- data.frame(x=runif(1000),y=runif(1000),g1=sample(letters[1:10],1000,replace=TRUE),g2=rep(letters[1:10],each=100),g3=factor(sample(letters[1:10],1000,replace=TRUE))) 
if(dbExistsTable(db,"test")) dbRemoveTable(db,"test") 
dbWriteTable(conn = db, name = "test", value = testDat, row.names=FALSE) 
# Read test 
testRecovery <- dbGetQuery(db, "SELECT * FROM test") 
testSelection <- dbGetQuery(db, "SELECT * FROM test WHERE g3=='h' OR g3=='e' ") 
# Close 
dbDisconnect(db) 

Answer 1

它看起來很簡單的對我說：factor是一個概念，只有S和R知道。句號。

因此，要讓它們進入數據庫並返回，您需要編寫映射器。要麼簡單化，要麼做所有的事情as.character（並且假設大部分數據庫後端會像R那樣散列字符串）。或者以數據庫爲中心，並將該因素分爲（無符號）（可能較短）整數和標籤。

Answer 2

好的，我按照@ DirkEddelbuettel的建議寫了一些包裝。評論贊賞。

#' Write a table via RSQLite with factors stored in another table 
#' Handles data.tables efficiently for large datasets 
#' @param conn The connection object (created with e.g. dbConnect) 
#' @param name The name of the table to write 
#' @param value The data.frame to write to the database 
#' @param factorName The base name of the tables to store the factor labels in in the SQLite database (e.g. if factorName is "_factor_" and the data.frame in value contains a factor column called "color" and the name is "mytable" then dbWriteFactorTable will create a table called mytable_factor_color which will store the levels information) 
#' @param \dots Options to pass along to dbWriteTable (e.g. append=TRUE) 
#' @return A boolean indicating whether the table write was successful 
dbWriteFactorTable <- function(conn, name, value, factorName="_factor_", ...) { 
    # Test inputs 
    stopifnot(class(conn)=="SQLiteConnection") 
    stopifnot(class(name)=="character") 
    stopifnot("data.frame" %in% class(value)) 
    stopifnot(class(factorName)=="character") 
    if(grepl("[.]",factorName)) stop("factorName must use valid characters for SQLite") 
    if("data.table" %in% class(value)) dt <- TRUE # Is value a data.table, if so use more efficient methods 
    # Convert factors to character 
    factorCols <- names(Filter(function(x) x=="factor", vapply(value, class, ""))) 
    if(length(factorCols>0)) { 
    for(cl in which(colnames(value) %in% factorCols)) { 
     cn <- colnames(value)[cl] 
     factorTable <- data.frame(levels=levels(value[[ cn ]])) 
     factorTable$levelKey <- seq(nrow(factorTable)) 
     fctNm <- paste0(name,factorName,cn) 
     dbWriteTable(conn = conn, name = fctNm, value = factorTable, row.names=FALSE, overwrite=TRUE) 
     if(dt) set(x=value, j=cl, value=as.character(value[[ cn ]])) 
    } 
    if(!dt) value <- japply(value, which(colnames(value) %in% factorCols), as.character) 
    } else { 
    warning("No factor columns detected.") 
    } 
    dbWriteTable(conn = conn, name = name, value = value, ...) 
} 

#' Read a table via RSQLite with factors stored in another table 
#' @param conn The connection object (created with e.g. dbConnect) 
#' @param name The name of the table to read 
#' @param query A character string containing sequel statements to be appended onto the query (e.g. "WHERE x==3") 
#' @param dt Whether to return a data.table vs. a plain-old data.frame 
#' @param factorName The base name of the tables to store the factor labels in in the SQLite database (e.g. if factorName is "_factor_" and the data.frame in value contains a factor column called "color" and the name is "mytable" then dbWriteFactorTable will expect there to be a table called mytable_factor_color which holds the levels information) 
#' @param \dots Options to pass along to dbGetQuery 
#' @return A data.table or data.frame 
dbReadFactorTable <- function(conn, name, query="", dt=TRUE, factorName="_factor_", ...) { 
    # Test inputs 
    stopifnot(class(conn)=="SQLiteConnection") 
    stopifnot(class(name)=="character") 
    stopifnot(class(factorName)=="character") 
    if(grepl("[.]",factorName)) stop("factorName must use valid characters for SQLite") 
    # Read main table 
    if(dt) { 
    value <- as.data.table(dbGetQuery(conn, paste("SELECT * FROM",name,query), ...)) 
    } else { 
    value <- dbGetQuery(conn, paste("SELECT * FROM",name,query), ...) 
    } 
    # Convert factors to character 
    factorCols <- sub(paste0("^.*",name,factorName,"(.+)$"), "\\1", 
    Filter(Negate(is.na), 
     str_extract(dbListTables(conn), paste0(".*",name,factorName,".*")) 
    ) 
) 
    if(length(factorCols>0)) { 
    for(cn in factorCols) { 
     fctNm <- paste0(name,factorName,cn) 
     factorTable <- dbGetQuery(conn, paste0("SELECT * FROM ",fctNm)) 
     if(dt) { 
     cl <- which(colnames(value) %in% cn) 
     set(x=value, j=cl, value=factor(value[[ cn ]], levels=factorTable$levels)) 
     } else { 
     value[[ cn ]] <- factor(value[[ cn ]], levels=factorTable$levels) 
     } 
    } 
    } else { 
    warning("No factor columns detected.") 
    } 
    value 
} 

而且一個簡單的例子：

db <- dbConnect(SQLite(), dbname="~/temp/test.sqlite") 
set.seed(1) 
n <- 1000 
testDat <- data.frame(key=seq(n), x=runif(n),y=runif(n),g1=sample(letters[1:10],n,replace=TRUE),g2=rep(letters[1:10],each=n/10),g3=factor(sample(letters[1:10],n,replace=TRUE))) 
if(dbExistsTable(db,"test")) dbRemoveTable(db,"test") 
dbWriteFactorTable(conn = db, name = "test", value = as.data.table(testDat), row.names=FALSE) 
dbReadFactorTable(conn = db, name = "test") 
dbReadFactorTable(conn = db, name = "test", query="WHERE g3=='a'")