csv - Pandas read_csv and UTF-16

Question

I have a CSV text file encoded in UTF-16 (so as to preserve Unicode characters when others use Excel) but when doing a read_csv with Pandas 0.9.0, I get this cryptic error:

df = pd.read_csv('data.txt',encoding='utf-16',sep='',header=0)
df.head()

---------------------------------------------------------------------------
Exception                                 Traceback (most recent call last)
<ipython-input-18-85da1383cd9e> in <module>()
----> 1 df = pd.read_csv('candidates-spanish.txt',encoding='utf-16',sep='',header=0)
  2 df.head()

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/io/parsers.pyc in read_csv(filepath_or_buffer, sep, dialect, header, index_col, names, skiprows, na_values, keep_default_na, thousands, comment, parse_dates, keep_date_col, dayfirst, date_parser, nrows, iterator, chunksize, skip_footer, converters, verbose, delimiter, encoding, squeeze, **kwds)
248         kdict['delimiter'] = sep
249 
--> 250     return _read(TextParser, filepath_or_buffer, kdict)
251 
252 @Appender(_read_table_doc)

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/io/parsers.pyc in _read(cls, filepath_or_buffer, kwds)
198         return parser
199 
--> 200     return parser.get_chunk()
201 
202 @Appender(_read_csv_doc)

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/io/parsers.pyc in get_chunk(self, rows)
853         elif not self._has_complex_date_col:
854             index = self._get_simple_index(alldata, columns)
--> 855             index = self._agg_index(index)
856 
857         elif self._has_complex_date_col:

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/io/parsers.pyc in _agg_index(self, index, try_parse_dates)
980                 arr, _ = _convert_types(arr, col_na_values)
981                 arrays.append(arr)
--> 982             index = MultiIndex.from_arrays(arrays, names=self.index_name)
983         return index
984 

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/core/index.pyc in from_arrays(cls, arrays, sortorder, names)
1570 
1571         return MultiIndex(levels=levels, labels=labels,
-> 1572                           sortorder=sortorder, names=names)
1573 
1574     @classmethod

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/core/index.pyc in __new__(cls, levels, labels, sortorder, names)
1254         assert(len(levels) == len(labels))
1255         if len(levels) == 0:
-> 1256             raise Exception('Must pass non-zero number of levels/labels')
1257 
1258         if len(levels) == 1:

Exception: Must pass non-zero number of levels/labels

Reading the data in line-by-line with csv.reader based on this example implies that my data is not incorrectly formatted:

from io import BytesIO
import csv

with open('data.txt','rb') as f:
    r = f.read().decode('utf-16').encode('utf-8')
    for l in csv.reader(BytesIO(r),delimiter=''):
        print l

['Country', 'State/City', 'Title', 'Date', 'Catalogue', 'Wikipedia Election Page', 'Wikipedia Individual Page', 'Electoral Institution in Country', 'Twitter', 'CANDIDATE NAME 1', 'CANDIDATE NAME 2']
['Venezuela', 'N/A', 'President', '10/7/12', 'Hugo Rafael Chavez Frias', 'Hugo Chxc3xa1vez', 'Hugo Chxc3xa1vez', 'Hugo Chavez', 'Hugo Chxc3xa1vez Frxc3xadas', 'Hugo Chavez', 'Hugo Chxc3xa1vez']
['Venezuela', 'N/A', 'President', '10/7/12', 'Henrique Capriles Radonski', 'Henrique Capriles Radonski', 'Henrique Capriles Radonski', 'Henrique Capriles Radonski', 'Henrique Capriles R.', 'Henrique Capriles', '']

Is there some pre-processing, an addition option in read_csv, or something else that needs to be done before pandas.read_csv can read a utf-16 file? Thanks!

Answer 1

This is a bug, I think because csv reader was passing back an extra empty line in the beginning. It worked for me on Python 2.7.3 and pandas 0.9.1 if I do:

In [36]: pd.read_csv(BytesIO(fh.read().decode('UTF-16').encode('UTF-8')), sep='', header=0)
Out[36]: 
<class 'pandas.core.frame.DataFrame'>
Int64Index: 50 entries, 0 to 49
Data columns:
Country                             43  non-null values
State/City                          43  non-null values
Title                               43  non-null values
Date                                43  non-null values
Catalogue                           43  non-null values
Wikipedia Election Page             43  non-null values
Wikipedia Individual Page           43  non-null values
Electoral Institution in Country    43  non-null values
Twitter                             43  non-null values
CANDIDATE NAME 1                    43  non-null values
CANDIDATE NAME 2                    16  non-null values
dtypes: object(11)

I reported the bug here: https://github.com/pydata/pandas/issues/2418 On github master it unfortunately causes a segfault in the c-parser. We'll fix it.

Now, interestingly: https://softwareengineering.stackexchange.com/questions/102205/should-utf-16-be-considered-harmful ;)