How do you split reading a large csv file into evenly-sized chunks in Python?

Question

In a basic I had the next process.

import csv
reader = csv.reader(open('huge_file.csv', 'rb'))

for line in reader:
    process_line(line)

See this related question. I want to send the process line every 100 rows, to implement batch sharding.

The problem about implementing the related answer is that csv object is unsubscriptable and can not use len.

>>> import csv
>>> reader = csv.reader(open('dataimport/tests/financial_sample.csv', 'rb'))
>>> len(reader)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: object of type '_csv.reader' has no len()
>>> reader[10:]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: '_csv.reader' object is unsubscriptable
>>> reader[10]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: '_csv.reader' object is unsubscriptable

How can I solve this?

Answer 1

Just make your reader subscriptable by wrapping it into a list. Obviously this will break on really large files (see alternatives in the Updates below):

>>> reader = csv.reader(open('big.csv', 'rb'))
>>> lines = list(reader)
>>> print lines[:100]
...

Further reading: How do you split a list into evenly sized chunks in Python?

---

Update 1 (list version): Another possible way would just process each chuck, as it arrives while iterating over the lines:

#!/usr/bin/env python

import csv
reader = csv.reader(open('4956984.csv', 'rb'))

chunk, chunksize = [], 100

def process_chunk(chuck):
    print len(chuck)
    # do something useful ...

for i, line in enumerate(reader):
    if (i % chunksize == 0 and i > 0):
        process_chunk(chunk)
        del chunk[:]
    chunk.append(line)

# process the remainder
process_chunk(chunk)

---

Update 2 (generator version): I haven't benchmarked it, but maybe you can increase performance by using a chunk generator:

#!/usr/bin/env python

import csv
reader = csv.reader(open('4956984.csv', 'rb'))

def gen_chunks(reader, chunksize=100):
    """ 
    Chunk generator. Take a CSV `reader` and yield
    `chunksize` sized slices. 
    """
    chunk = []
    for i, line in enumerate(reader):
        if (i % chunksize == 0 and i > 0):
            yield chunk
            del chunk[:]
        chunk.append(line)
    yield chunk

for chunk in gen_chunks(reader):
    print chunk # process chunk

# test gen_chunk on some dummy sequence:
for chunk in gen_chunks(range(10), chunksize=3):
    print chunk # process chunk

# => yields
# [0, 1, 2]
# [3, 4, 5]
# [6, 7, 8]
# [9]

Answer 2

There isn't a good way to do this for all .csv files. You should be able to divide the file into chunks using file.seek to skip a section of the file. Then you have to scan one byte at a time to find the end of the row. The you can process the two chunks independently. Something like the following (untested) code should get you started.

file_one = open('foo.csv')
file_two = open('foo.csv') 
file_two.seek(0, 2)     # seek to the end of the file
sz = file_two.tell()    # fetch the offset
file_two.seek(sz / 2)   # seek back to the middle
chr = ''
while chr != '\n':
    chr = file_two.read(1)
# file_two is now positioned at the start of a record
segment_one = csv.reader(file_one)
segment_two = csv.reader(file_two)

I'm not sure how you can tell that you have finished traversing segment_one. If you have a column in the CSV that is a row id, then you can stop processing segment_one when you encounter the row id from the first row in segment_two.

Answer 3

We can use pandas module to handle these big csv files.

df = pd.DataFrame()
temp = pd.read_csv('BIG_File.csv', iterator=True, chunksize=1000)
df = pd.concat(temp, ignore_index=True)