Sorting CSV in Python

Question

I assumed sorting a CSV file on multiple text/numeric fields using Python would be a problem that was already solved. But I can't find any example code anywhere, except for specific code focusing on sorting date fields.

How would one go about sorting a relatively large CSV file (tens of thousand lines) on multiple fields, in order?

Python code samples would be appreciated.

Answer 1

Python's sort works in-memory only; however, tens of thousands of lines should fit in memory easily on a modern machine. So:

import csv

def sortcsvbymanyfields(csvfilename, themanyfieldscolumnnumbers):
  with open(csvfilename, 'rb') as f:
    readit = csv.reader(f)
    thedata = list(readit)
  thedata.sort(key=operator.itemgetter(*themanyfieldscolumnnumbers))
  with open(csvfilename, 'wb') as f:
    writeit = csv.writer(f)
    writeit.writerows(thedata)

Answer 2

You bring up 3 issues:

• file size
• csv data
• sorting on multiple fields

Here is a solution for the third part. You can handle csv data in a more sophisticated way.

>>> data = 'a,b,c\nb,b,a\nb,c,a\n'
>>> lines = [e.split(',') for e in data.strip().split('\n')]
>>> lines
[['a', 'b', 'c'], ['b', 'b', 'a'], ['b', 'c', 'a']]
>>> def f(e):
...     field_order = [2,1]
...     return [e[i] for i in field_order]
... 
>>> sorted(lines, key=f)
[['b', 'b', 'a'], ['b', 'c', 'a'], ['a', 'b', 'c']]

Edited to use a list comprehension, generator does not work as I had expected it to.

Answer 3

Here's Alex's answer, reworked to support column data types:

import csv
import operator

def sort_csv(csv_filename, types, sort_key_columns):
    """sort (and rewrite) a csv file.
    types:  data types (conversion functions) for each column in the file
    sort_key_columns: column numbers of columns to sort by"""
    data = []
    with open(csv_filename, 'rb') as f:
        for row in csv.reader(f):
            data.append(convert(types, row))
    data.sort(key=operator.itemgetter(*sort_key_columns))
    with open(csv_filename, 'wb') as f:
        csv.writer(f).writerows(data)

Edit:

I did a stupid. I was playing with various things in IDLE and wrote a convert function a couple of days ago. I forgot I'd written it, and I haven't closed IDLE in a good long while - so when I wrote the above, I thought convert was a built-in function. Sadly no.

Here's my implementation, though John Machin's is nicer:

def convert(types, values):
    return [t(v) for t, v in zip(types, values)]

Usage:

import datetime
def date(s):
    return datetime.strptime(s, '%m/%d/%y')

>>> convert((int, date, str), ('1', '2/15/09', 'z'))
[1, datetime.datetime(2009, 2, 15, 0, 0), 'z']

Answer 4

Here's the convert() that's missing from Robert's fix of Alex's answer:

>>> def convert(convert_funcs, seq):
...    return [
...        item if func is None else func(item)
...        for func, item in zip(convert_funcs, seq)
...        ]
...
>>> convert(
...     (None, float, lambda x: x.strip().lower()),
...     [" text ", "123.45", " TEXT "]
...     )
[' text ', 123.45, 'text']
>>>

I've changed the name of the first arg to highlight that the per-columns function can do what you need, not merely type-coercion. None is used to indicate no conversion.