pandas unable to read from large StringIO object

Question

I'm using pandas to manage a large array of 8-byte integers. These integers are included as space-delimited elements of a column in a comma-delimited CSV file, and the array size is about 10000x10000.

Pandas is able to quickly read the comma-delimited data from the first few columns as a DataFrame, and also quickly store the space-delimited strings in another DataFrame with minimal hassle. The trouble comes when I try to cast transform the table from a single column of space-delimited strings to a DataFrame of 8-bit integers.

I have tried the following:

intdata = pd.DataFrame(strdata.columnname.str.split().tolist(), dtype='uint8')

But the memory usage is unbearable - 10MB worth of integers consumes 2GB of memory. I'm told that it's a limitation of the language and there's nothing I can do about it in this case.

As a possible workaround, I was advised to save the string data to a CSV file and then reload the CSV file as a DataFrame of space-delimited integers. This works well, but to avoid the slowdown that comes from writing to disk, I tried writing to a StringIO object.

Here's a minimal non-working example:

import numpy as np
import pandas as pd
from cStringIO import StringIO

a = np.random.randint(0,256,(10000,10000)).astype('uint8')
b = pd.DataFrame(a)
c = StringIO()
b.to_csv(c, delimiter=' ', header=False, index=False)
d = pd.io.parsers.read_csv(c, delimiter=' ', header=None, dtype='uint8')

Which yields the following error message:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib64/python2.7/site-packages/pandas/io/parsers.py", line 443, in parser_f
    return _read(filepath_or_buffer, kwds)
  File "/usr/lib64/python2.7/site-packages/pandas/io/parsers.py", line 228, in _read
    parser = TextFileReader(filepath_or_buffer, **kwds)
  File "/usr/lib64/python2.7/site-packages/pandas/io/parsers.py", line 533, in __init__
    self._make_engine(self.engine)
  File "/usr/lib64/python2.7/site-packages/pandas/io/parsers.py", line 670, in _make_engine
    self._engine = CParserWrapper(self.f, **self.options)
  File "/usr/lib64/python2.7/site-packages/pandas/io/parsers.py", line 1032, in __init__
    self._reader = _parser.TextReader(src, **kwds)
  File "parser.pyx", line 486, in pandas.parser.TextReader.__cinit__ (pandas/parser.c:4494)
ValueError: No columns to parse from file

Which is puzzling, because if I run the exact same code with 'c.csv' instead of c, the code works perfectly. Also, if I use the following snippet:

file = open('c.csv', 'w')
file.write(c.getvalue())

The CSV file gets saved without any problems, so writing to the StringIO object is not the issue.

It is possible that I need to replace c with c.getvalue() in the read_csv line, but when I do that, the interpreter tries to print the contents of c in the terminal! Surely there is a way to work around this.

Thanks in advance for the help.

Answer 1

There are two issues here, one fundamental and one you simply haven't come across yet. :^)

First, after you write to c, you're at the end of the (virtual) file. You need to seek back to the start. We'll use a smaller grid as an example:

>>> a = np.random.randint(0,256,(10,10)).astype('uint8')
>>> b = pd.DataFrame(a)
>>> c = StringIO()
>>> b.to_csv(c, delimiter=' ', header=False, index=False)
>>> next(c)
Traceback (most recent call last):
  File "<ipython-input-57-73b012f9653f>", line 1, in <module>
    next(c)
StopIteration

which generates the "no columns" error. If we seek first, though:

>>> c.seek(0)
>>> next(c)
'103,3,171,239,150,35,224,190,225,57\n'

But now you'll notice the second issue-- commas? I thought we requested space delimiters? But to_csv only accepts sep, not delimiter. Seems to me it should either accept it or object that it doesn't, but silently ignoring it feels like a bug. Anyway, if we use sep (or delim_whitespace=True):

>>> a = np.random.randint(0,256,(10,10)).astype('uint8')
>>> b = pd.DataFrame(a)
>>> c = StringIO()
>>> b.to_csv(c, sep=' ', header=False, index=False)
>>> c.seek(0)
>>> d = pd.read_csv(c, sep=' ', header=None, dtype='uint8')
>>> d
     0    1    2    3    4    5    6    7    8    9
0  209   65  218  242  178  213  187   63  137  145
1  161  222   50   92  157   31   49   62  218   30
2  182  255  146  249  115   91  160   53  200  252
3  192  116   87   85  164   46  192  228  104  113
4   89  137  142  188  183  199  106  128  110    1
5  208  140  116   50   66  208  116   72  158  169
6   50  221   82  235   16   31  222    9   95  111
7   88   36  204   96  186  205  210  223   22  235
8  136  221   98  191   31  174   83  208  226  150
9   62   93  168  181   26  128  116   92   68  153