reading csv files in scipy/numpy in Python

Question

I am having trouble reading a csv file, delimited by tabs, in python. I use the following function:

def csv2array(filename, skiprows=0, delimiter='\t', raw_header=False, missing=None, with_header=True):
    """
    Parse a file name into an array. Return the array and additional header lines. By default,
    parse the header lines into dictionaries, assuming the parameters are numeric,
    using 'parse_header'.
    """
    f = open(filename, 'r')
    skipped_rows = []
    for n in range(skiprows):
        header_line = f.readline().strip()
        if raw_header:
            skipped_rows.append(header_line)
        else:
            skipped_rows.append(parse_header(header_line))
    f.close()
    if missing:
        data = genfromtxt(filename, dtype=None, names=with_header,
                          deletechars='', skiprows=skiprows, missing=missing)
    else:
    if delimiter != '\t':
        data = genfromtxt(filename, dtype=None, names=with_header, delimiter=delimiter,
                  deletechars='', skiprows=skiprows)
    else:
        data = genfromtxt(filename, dtype=None, names=with_header,
                  deletechars='', skiprows=skiprows)        
    if data.ndim == 0:
    data = array([data.item()])
    return (data, skipped_rows)

the problem is that genfromtxt complains about my files, e.g. with the error:

Line #27100 (got 12 columns instead of 16)

I am not sure where these errors come from. Any ideas?

Here's an example file that causes the problem:

#Gene   120-1   120-3   120-4   30-1    30-3    30-4    C-1 C-2 C-5 genesymbol  genedesc
ENSMUSG00000000001  7.32    9.5 7.76    7.24    11.35   8.83    6.67    11.35   7.12    Gnai3   guanine nucleotide binding protein alpha
ENSMUSG00000000003  0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Pbsn    probasin

Is there a better way to write a generic csv2array function? thanks.

Solution 1

Check out the python CSV module: http://docs.python.org/library/csv.html

import csv
reader = csv.reader(open("myfile.csv", "rb"), 
                    delimiter='\t', quoting=csv.QUOTE_NONE)

header = []
records = []
fields = 16

if thereIsAHeader: header = reader.next()

for row, record in enumerate(reader):
    if len(record) != fields:
        print "Skipping malformed record %i, contains %i fields (%i expected)" %
            (record, len(record), fields)
    else:
        records.append(record)

# do numpy stuff.

Solution 2

May I ask why you're not using the built-in csv reader? http://docs.python.org/library/csv.html

I've used it very effectively with numpy/scipy. I would share my code but unfortunately it's owned by my employer, but it should be very straightforward to write your own.

Comments

[Admin]

I added an example file that leads to the error -- it looks to me like it has the right number of columns but for some reason it thinks it has 16 columns. Any idea what causes this?

[Admin]

this does not make a numpy array out of the result, unfortunately

[Nick T]

You can do whatever you like with the data in the loop body; there it's a list broken up by delimiter. You could check if it's as long as you expect, (in edited example), or do validation on each field to make sure you're not passing garbage into your numpy array.