Delete Multiple Columns From 500 Mb Tsv File With Python (or Perl Etc)

I have a very large tsv file and need to delete several columns. I've found the CSV module, and an answer as below to a sort of similar question (see script below). Yet I need to d

Solution 1:

You can use del to delete slices of a list.

withopen('in.tsv', 'r') as fin, open('out.tsv', 'w') as fout:
    reader = csv.reader(fin, dialect='excel-tab')
    writer = csv.writer(fout, dialect='excel-tab')
    for row in reader:
        # delete indices in reverse order to avoid shifting earlier indicesdel row[653321:689513+1]
        del row[628715:650181+1]
        writer.writerow(row)

Solution 2:

You can do this with very little memory using Python.

First define a dialect describing your tsv format. See the documentation on dialects for more information.

classTsvDialect(csv.Dialect):
    delimiter = '\t'
    quoting = csv.QUOTE_NONE
    escapechar = None

# you can just pass thisclassaround, or you can register it under a name
csv.register_dialect('tsv', TsvDialect)

Then you can walk through each line and copy to a new tsv:

withopen('source.tsv', 'rb') as src, open('result.tsv', 'wb') as res:
    csrc = csv.reader(src, dialect='tsv')
    cres = csv.writer(res, dialect='tsv')
    for row in csrc:
        cres.writerow(row)

This does a simple copy. Since you only want some rows, lets only copy those.

Python's lists are zero-indexed (the first column is column 0, not column 1); and index slicing does not include the last item (wholelist[:2] is the same as [wholelist[0], wholelist[1]]). Keep these in mind to avoid off-by-one errors!

withopen('source.tsv', 'rb') as src, open('result.tsv', 'wb') as res:
    csrc = csv.reader(src, dialect='tsv')
    cres = csv.writer(res, dialect='tsv')
    for row in csrc:
        # remove [628714:650181] and [653320:689512]
        newrow = row[:628714] # columns before 628714
        newrow.extend(row[650181:653320]) # columns between 650180 and 653320
        cres.writerow(newrow)

Alternatively, instead of copying the columns you want to a new row, you can save some memory at the expense of code clarity by deleting the columns you don't want:

forrowincsrc:
        # remove[628714:650181]and[653320:689512]
        # besuretoremoveinreverseorder!
        delrow[653320:689512]delrow[628714:650181]cres.writerow(row)

You can abstract column cutting (either method, using any indexing you're comfortable with) into a function if you need to do this very often.

You might also want to take a look at the csvkit python library and command-line tools, in particular its command-line tool csvcut, which appears to do exactly what you want from the command line.

Solution 3:

With 2 GB RAM or more, it should be possible to load the dataset in memory, delete the columns you want, and write the contents to a file. This could either be done in R or python easily. For R:

dat = read.table("spam.tsv", ...)
dat = dat[-c(1,5)] # delete row 1and5write.csv(dat, ....)

Doing this in chunks can easily be done using either an apply loop or a for loop. I use the apply style:

read_chunk = function(chunk_index, chunk_size, fname) {
    dat = read.table(fname, nrow = chunk_size, skip = (chunk_id - 1) * chunk_size, ...)
    dat = dat[-c(1,5)] # delete row 1 and 5
    write.csv(dat, append = TRUE, ....)    
}

tot_no_lines = 10000 # for example
chunk_size = 1000
sapply(1:(tot_no_lines / chunk_size), read_chunk)

Note that this is R style code useful as inspiration, no working R code.

Solution 4:

You can build the output row dynamically:

for r in rdr:
    outrow = []
    for i in range(0, 628714):
       outrow.append(r[i])
    for i in range(650181, 653320):
       outrow.append(r[i])
    wtr.writerow( outrow )

I imagine you can do this even more concisely with slices of the input row r, along the lines of:

outrow = r[0:628714)
 outrow.extend(r[650181:653320)
 wrt.writerow( outrow )

Perhaps not the fastest to execute, but certainly easier to write.

Solution 5:

Are you on Linux? Then save the hazzle and use csvtool from shell:

 csvtool col 1-500,502-1000input.csv > output.csv

You can also set delimiter and so on, just type csvtool --help. Quite easy to use.