How to Handle Large Datasets in R - Part 2

Master R

By Guangming Lang Comment

R is my favorite tool for data analysis except when it comes to dealing with large datasets. If the data file is larger than the size of your RAM, R will fail at reading it in. So when you have a large dataset, you should first check if you have enough memory. I showed a method of doing that when all you know is the number of rows and columns in the data file.

Let’s say you find out the data file is too large for R to handle, what would you do? You can take a random sample from the data file, read and analyze the sample using R. The following readbig() function does that. It reads the input data file in chunks, takes a sample of user specified size, and returns the read-in data as a data frame.

readbig = function (file, samplesz, chunksz, sep=',', header=TRUE, nrec=0, ...) {
        ## this function reads in a random sample of a big input data file
        ## @file: input file path
        ## @samplesz: sample size
        ## @chunksz: size of chunks to read in at each iteration
        ## @header: logical value indicating if the input file has a header
        ## @nrec: number of rows in input file. If @nrec is <= 0, this function
        ##        will use operating system command to find out its value.
        ##        Default value is 0. 
        # a pretty efficient way to find the number of lines in file
        # comment out the following appropriate codeblocks (line16-28) 
        # depending on your operating system.
#         # Windows: 
#         # shell('type "comma.txt" | find /c ","', intern=TRUE)
#         if(nrec <= 0) 
#                 nrec = as.numeric(
#                         shell( paste0('type ',file,' | find /c ','"',sep,'"'), 
#                                intern=TRUE ) )
        # Mac or Linux:
        # system('cat comma.txt | wc -l', intern=TRUE)
        if(nrec <= 0) 
                nrec = as.numeric(
                        system( paste0('cat ',file,' | wc -l'), 
                                intern=TRUE ) )
        # create a file connection
        f = file(file, 'r')
        # take a sample (of size samplesz) of the row indices and sort them
        use = sort(sample(1:nrec, samplesz))
        # read the 1st line
        line1 = readLines(f,1) 
        # calculate the number of cols of the file
        line1 = unlist(strsplit(line1,sep)[[1]])
        k = length(line1)
        # re-position the file to its origin
        # make a zero-matrix with samplesz rows and k cols
        result = data.frame(matrix(NA, samplesz, k))
        # initialize some values
        read = 0
        got = 1
        left = nrec
        skip_row = 0
        # skip the header if there's one
        if (header) {
                left = nrec - 1 
                skip_row = 1
                names(result) = line1
        while(left > 0){
                # read the next chunk (each chuck has size chunksz) from f
                now = read.table(f, nrows=chunksz, skip=skip_row, sep=sep, ...) 
                # don't skip the 1st row when reading in the 2nd, 3rd, ... block
                skip_row = 0   
                begin = read + 1
                end = read + chunksz
                # extract row indices in both the chunk and the sample 
                want = (begin:end)[begin:end %in% use] - read 
                # get sampled data
                if (length(want) > 0) {
                        nowdat = now[want,]
                        newgot = got + length(want) - 1
                        result[got:newgot,] = nowdat
                        got = newgot + 1
                read = read + chunksz
                left = left - chunksz

If you enjoyed this post, get updates. It's FREE