如何计算分隔文件中的非空字段数？

Question

You can count the number of fields per line in a comma/tab/whatever delimited text file using utils::count.fields .

Here's a reproducible example:

d <- data.frame(
  x = c(1, NA, 3, NA, 5),
  y = c(NA, "b", "c", NA, NA),
  z = c(NA, "beta", "gamma", NA, "epsilon")
)

fname <- "test.csv"
write.csv(d, fname, na = "",  row.names = FALSE)
count.fields(fname, sep = ",")
## [1] 3 3 3 3 3 3

I want to calculate the number of non-empty fields per line. I can do this in a clunky way by reading in everything and counting the number of values that aren't NA .

d2 <- read.csv(fname, na.strings = "")
rowSums(!is.na(d2))
## [1] 1 2 3 0 2

I'd really like a way of scanning the file (like count.fields ) so I can target specific sections to read in.

Is there a better way of counting the number of non-empty fields in a delimited file?

Answer 1

This should be completely portable provided you have the Rcpp & BH packages installed:

library(Rcpp)
library(inline)

csvblanks <- '
string data = as<string>(filename);
ifstream fil(data.c_str());
if (!fil.is_open()) return(R_NilValue);

typedef tokenizer< escaped_list_separator<char> > Tokenizer;

vector<string> fields;
vector<int> retval;
string line;

while (getline(fil, line)) {
  int numblanks = 0;
  Tokenizer tok(line);
  for(Tokenizer::iterator beg=tok.begin(); beg!=tok.end(); ++beg){
    numblanks += (beg->length() == 0) ? 1 : 0 ;
  };
  retval.push_back(numblanks);
}
return(wrap(retval));
'

count_blanks <- rcpp(
  signature(filename="character"),
  body=csvblanks,
  includes=c("#include <iostream>",
             "#include <fstream>",
             "#include <vector>",
             "#include <string>",
             "#include <algorithm>",
             "#include <iterator>",
             "#include <boost/tokenizer.hpp>",
             "using namespace Rcpp;",
             "using namespace std;",
             "using namespace boost;")
)

Once that's sourced you can call count_blanks(FULLPATH) and it will return a numeric vector of counts of blank fields per line.

I ran it against this file:

"DATE","APIKEY","FILENAME","LANGUAGE","JOBID","TRANSCRIPT"
1,2,3,4,5
1,,3,4,5
1,2,3,4,5
1,2,,4,5
1,2,3,4,5
1,2,3,,5
1,2,3,4,5
1,2,3,4,
1,2,3,4,5
1,,3,,5
1,2,3,4,5
,2,,4,
1,2,3,4,5

via:

count_blanks("/tmp/a.csv")
## [1] 0 0 1 0 1 0 1 0 1 0 2 0 3 0

CAVEATS

• It's fairly obvious that it's not ignoring the header, so it could use a header logical parameter with associated C/C++ code (which will be pretty straightforward).
• If you're counting "spaces" (ie [:space:]+ ) as "empty" you'll need something a bit more complex than the call to length . This is one potential way to deal with it if you need to.
• It's using the default configuration for the Boost function escaped_list_separator which is defined here . That can also be customized with with quote & separator characters (making it possible to further mimic read.csv / read.table .

This will more closely approach count.fields / C_countfields performance and will eliminate the need to consume memory by reading in every line just to find the lines you eventually want to more optimally target. I don't think preallocating space for the returned numeric vector will add much to the speed, but you can see the discussion here which shows how to do so if need be.