在R中导入大型csv文件，read.csv.ffdf中出现错误

Question

I want to import a faily large file (40Mrows x 4columns). I ended up using ffbase , after a try to sqldf

I tried base::read.csv : It failed. I tried sqldf::sqldf : It failed too saying it could not allocate anymore.

I am just trying to replicate the example given in the ffbase vignette.

R) x <- data.frame(log=rep(c(FALSE, TRUE), length.out=26), int=1:26, dbl=1:26 + 0.1,   fac=factor(letters), ord=ordered(LETTERS), dct=Sys.time()+1:26, dat=seq(as.Date("1910/1/1"), length.out=26, by=1))
R) x <- x[c(13:1, 13:1),]
R) csvfile <- tempPathFile(path=getOption("fftempdir"), extension="csv")
R) write.csv(x, file=csvfile, row.names=FALSE)
R) y <- read.csv(file=csvfile, header=TRUE)
R) y
 log int  dbl fac ord                       dct        dat
1  FALSE  13 13.1   m   M 2012-11-26 11:21:29.15763 1910-01-13
2   TRUE  12 12.1   l   L 2012-11-26 11:21:28.15763 1910-01-12
3  FALSE  11 11.1   k   K 2012-11-26 11:21:27.15763 1910-01-11
4   TRUE  10 10.1   j   J 2012-11-26 11:21:26.15763 1910-01-10
...
23  TRUE   4  4.1   d   D 2012-11-26 11:21:20.15763 1910-01-04
24 FALSE   3  3.1   c   C 2012-11-26 11:21:19.15763 1910-01-03
25  TRUE   2  2.1   b   B 2012-11-26 11:21:18.15763 1910-01-02
26 FALSE   1  1.1   a   A 2012-11-26 11:21:17.15763 1910-01-01


# ---- !!!!! HERE !!!! ---- #
R) ffx <- read.csv.ffdf(file=csvfile, header=TRUE)
Erreur dans ff(initdata = initdata, length = length, levels = levels, ordered = ordered,  : vmode 'character' not implemented

I don't understand...

Do you have any insight?

Answer 1

You probably need to pass the argument colClasses as follows. As you would do with a normal read.csv

ffx <- read.csv.ffdf(file=csvfile, header=TRUE, colClasses = c("logical","integer","numeric","factor","factor","POSIXct","Date"))

Answer 2

sorry I am late I had no access to R last 3 days. Here is some additional code for read.csv

  R) setAs("character","myDate", function(from) as.Date(from, format="%d/%m/%y") )
  R) system.time(data <- read.csv(file=filePath, sep=";", stringsAsFactors=TRUE, colClasses=c("factor","factor","numeric","myDate"), nrows=10));

    utilisateur     système      écoulé 
    0               0            0 
  R) system.time(data <- read.csv(file=filePath, sep=";", stringsAsFactors=TRUE, colClasses=c("factor","factor","numeric","myDate")));
    Erreur : impossible d'allouer un vecteur de taille 250.0 Mo
    Timing stopped at: 236.2 4.92 333.3 

=> So read.csv can't handle that number of lines.

---

Same test for read.csv.sql which is a wrapper of sqldf only for 500 rows.

R) system.time(data <- read.csv.sql(filePath, dbname = tempfile(), header = T, row.names = F, sep=";"));
   utilisateur     système      écoulé 
   0.07            0.00        0.07 

BTW please be advise that the nbrows option is !NOT WORKING!... abd that you cannot indicate a colClasses argument...

R) system.time(data <- sqldf("select * from f", dbname = tempfile(), file.format = list(header = T, row.names = F, sep=";")));
   Erreur : impossible d'allouer un vecteur de taille 500.0 Mo
   Timing stopped at: 366.8 42.45 570.2

For the whole table it crashes... Strange as it is supposed to be a reference for big data...

---

And finally using package ff , for 50 rows

R) system.time(data <- read.csv.ffdf(file=filePath, header=TRUE, nrows=50, colClasses=c("factor","factor","numeric","myDate"),sep=";"))
   utilisateur     système      écoulé 
   0.02            0.00         0.03 

Please note that head(data) also has a bug, it does not display columns accurately...

And for the whole table... IT WORKS ... !fireworks!

R) system.time(data <- read.csv.ffdf(file=filePath, header=TRUE, colClasses=c("factor","factor","numeric","myDate"),sep=";"))
   utilisateur     système      écoulé 
   409.69          14.42        547.75 

For a 36M rows table

R) dim(data)
   [1] 36083010        4

---

As a consequence I recommand ff package to load big dataset