automatically detect date columns when reading a file into a data.frame

Question

When reading a file, the read.table function uses type.convert to distinguish between logical, integer, numeric, complex, or factor columns and store them accordingly.

I'd like to add dates to the mix, so that columns containing dates can automatically be recognized and parsed into Date objects. Only a few date formats should be recognized, e.g.

date.formats <- c("%m/%d/%Y", "%Y/%m/%d")

Here is an example:

fh <- textConnection(

 "num  char date-format1  date-format2  not-all-dates  not-same-formats
   10     a     1/1/2013    2013/01/01     2013/01/01          1/1/2013
   20     b     2/1/2013    2013/02/01              a        2013/02/01 
   30     c     3/1/2013            NA              b          3/1/2013"
)

And the output of

dat <- my.read.table(fh, header = TRUE, stringsAsFactors = FALSE,
                     date.formats = date.formats)
sapply(dat, class)

would give:

num              => numeric
char             => character
date-format1     => Date
date-format2     => Date
not-all-dates    => character
not-same-formats => character   # not a typo: date format must be consistent

Before I go and implement it from scratch, is something like this already available in a package? Or maybe someone already gave it a crack (or will) and is willing to share his code here? Thank you.

Answer 1

You could use lubridate::parse_date_time, which is a bit stricter (and creates POSIXlt) data.

I've also added a bit more checking for existing NA values (may not be necessary).

eg

library(lubridate)
my.read.table <- function(..., date.formats = c("%m/%d/%Y", "%Y/%m/%d")) {
  dat <- read.table(...)
  for (col.idx in seq_len(ncol(dat))) {
    x <- dat[, col.idx]
    if(!is.character(x) | is.factor(x)) next
    if (all(is.na(x))) next
    for (format in date.formats) {
      complete.x <- !(is.na(x))
      d <- as.Date(parse_date_time(as.character(x), format, quiet = TRUE))
      d.na <- d[complete.x]
      if (any(is.na(d.na))) next
      dat[, col.idx] <- d         
    }
  }
  dat

}

 dat <- my.read.table(fh, stringsAsFactors = FALSE,header=TRUE)

str(dat)
'data.frame':   3 obs. of  6 variables:
 $ num             : int  10 20 30
 $ char            : chr  "a" "b" "c"
 $ date.format1    : Date, format: "2013-01-01" "2013-02-01" "2013-03-01"
 $ date.format2    : Date, format: "2013-01-01" "2013-02-01" NA
 $ not.all.dates   : chr  "2013/01/01" "a" "b"
 $ not.same.formats: chr  "1/1/2013" "2013/02/01" "3/1/2013"

An alternative would be to use options(warn = 2) within the function and wrap the parse_date_time(...) in a try statement

my.read.table <- function(..., date.formats = c("%m/%d/%Y", "%Y/%m/%d")) {
  dat <- read.table(...)
  owarn <-getOption('warn')
  on.exit(options(warn = owarn))
  options(warn = 2)
  for (col.idx in seq_len(ncol(dat))) {
    x <- dat[, col.idx]
    if(!is.character(x) | is.factor(x)) next
    if (all(is.na(x))) next
    for (format in date.formats) {
      d <- try(as.Date(parse_date_time(as.character(x), format)), silent= TRUE)

      if (inherits(d, 'try-error')) next
      dat[, col.idx] <- d         
    }
  }
  dat

}

Answer 2

You can try with regular expressions.

my.read.table <- function(..., date.formats = c("%m/%d/%Y", "%Y/%m/%d")) {
   require(stringr)
   formats <- c(
     "%m" = "[0-9]{1,2}",
     "%d" = "[0-9]{1,2}",
     "%Y" = "[0-9]{4}"
   )
   dat <- read.table(...)
   for (col.idx in seq_len(ncol(dat))) {
      for (format in date.formats) {
         x <- dat[, col.idx]
         if(!is.character(x) | is.factor(x)) break
         if (all(is.na(x))) break
         x <- as.character(x)
         # Convert the format into a regular expression
         for( k in names(formats) ) {
           format <- str_replace_all( format, k, formats[k] )
         }
         # Check if it matches on the non-NA elements
         if( all( str_detect( x, format ) | is.na(x) ) ) {
           dat[, col.idx] <- as.Date(x, format)
           break
         }
      }
   }
   dat
}

dat <- my.read.table(fh, header = TRUE, stringsAsFactors = FALSE)
as.data.frame(sapply(dat, class))
#                  sapply(dat, class)
# num                         integer
# char                      character
# date.format1                   Date
# date.format2                   Date
# not.all.dates             character
# not.same.formats          character

Answer 3

Here I threw one together quickly. It is not handling the last column properly because the as.Date function is not strict enough (see that as.Date("1/1/2013", "%Y/%m/%d") parses ok for example...)

my.read.table <- function(..., date.formats = c("%m/%d/%Y", "%Y/%m/%d")) {
   dat <- read.table(...)
   for (col.idx in seq_len(ncol(dat))) {
      x <- dat[, col.idx]
      if(!is.character(x) | is.factor(x)) next
      if (all(is.na(x))) next
      for (f in date.formats) {
         d <- as.Date(as.character(x), f)
         if (any(is.na(d[!is.na(x)]))) next
         dat[, col.idx] <- d         
      }
   }
   dat
}

dat <- my.read.table(fh, header = TRUE, stringsAsFactors = FALSE)
as.data.frame(sapply(dat, class))

#                  sapply(dat, class)
# num                         integer
# char                      character
# date.format1                   Date
# date.format2                   Date
# not.all.dates             character
# not.same.formats               Date

If you know a way to parse dates that is more strict around formats than as.Date (see the example above), please let me know.

Edit: To make the date parsing super strict, I can add

if (!identical(x, format(d, f))) next

For it to work, I will need all my input dates to have leading zeroes where needed, i.e. 01/01/2013 and not 1/1/2013. I can live with that if that's the standard way.