ENH: Support high precision converters in to_numeric

[xmduhan]

import pandas as pd
from pandas.compat import StringIO
data = StringIO("""
a, b
1, 1
1.7976931348623157e+308, 1
""")
pd.read_csv(data, dtype={'a': float, 'b': float}, engine='python')

This work!

import pandas as pd
from pandas.compat import StringIO
data = StringIO("""
a, b
1, 1
1.7976931348623157e+308, 1
""")
pd.read_csv(data, dtype={'a': float, 'b': float}, engine='c')

This fail!
Message: ValueError: cannot safely convert passed user dtype of float64 for object dtyped data in column 0

Full error stack:

ValueErrorTraceback (most recent call last)
<ipython-input-4-f830b6441439> in <module>()
      5 1.7976931348623157e+308, 1
      6 """)
----> 7 pd.read_csv(data, dtype={'a': float, 'b': float}, engine='c')

/usr/local/lib/python2.7/dist-packages/pandas/io/parsers.pyc in parser_f(filepath_or_buffer, sep, delimiter, header, names, index_col, usecols, squeeze, prefix, mangle_dupe_cols, dtype, engine, converters, true_values, false_values, skipinitialspace, skiprows, nrows, na_values, keep_default_na, na_filter, verbose, skip_blank_lines, parse_dates, infer_datetime_format, keep_date_col, date_parser, dayfirst, iterator, chunksize, compression, thousands, decimal, lineterminator, quotechar, quoting, escapechar, comment, encoding, dialect, tupleize_cols, error_bad_lines, warn_bad_lines, skipfooter, skip_footer, doublequote, delim_whitespace, as_recarray, compact_ints, use_unsigned, low_memory, buffer_lines, memory_map, float_precision)
    707                     skip_blank_lines=skip_blank_lines)
    708 
--> 709         return _read(filepath_or_buffer, kwds)
    710 
    711     parser_f.__name__ = name

/usr/local/lib/python2.7/dist-packages/pandas/io/parsers.pyc in _read(filepath_or_buffer, kwds)
    453 
    454     try:
--> 455         data = parser.read(nrows)
    456     finally:
    457         parser.close()

/usr/local/lib/python2.7/dist-packages/pandas/io/parsers.pyc in read(self, nrows)
   1067                 raise ValueError('skipfooter not supported for iteration')
   1068 
-> 1069         ret = self._engine.read(nrows)
   1070 
   1071         if self.options.get('as_recarray'):

/usr/local/lib/python2.7/dist-packages/pandas/io/parsers.pyc in read(self, nrows)
   1837     def read(self, nrows=None):
   1838         try:
-> 1839             data = self._reader.read(nrows)
   1840         except StopIteration:
   1841             if self._first_chunk:

pandas/_libs/parsers.pyx in pandas._libs.parsers.TextReader.read()

pandas/_libs/parsers.pyx in pandas._libs.parsers.TextReader._read_low_memory()

pandas/_libs/parsers.pyx in pandas._libs.parsers.TextReader._read_rows()

pandas/_libs/parsers.pyx in pandas._libs.parsers.TextReader._convert_column_data()

pandas/_libs/parsers.pyx in pandas._libs.parsers.TextReader._convert_tokens()

ValueError: cannot safely convert passed user dtype of float64 for object dtyped data in column 0

### pd.show_versions(): INSTALLED VERSIONS ------------------ commit: None python: 2.7.6.final.0 python-bits: 64 OS: Linux OS-release: 3.13.0-139-generic machine: x86_64 processor: x86_64 byteorder: little LC_ALL: None LANG: zh_CN.UTF-8 LOCALE: None.None

pandas: 0.22.0
pytest: None
pip: 9.0.1
setuptools: 38.4.0
Cython: 0.27.3
numpy: 1.14.0
scipy: 1.0.0
pyarrow: None
xarray: None
IPython: 5.5.0
sphinx: 1.6.6
patsy: 0.5.0
dateutil: 2.6.1
pytz: 2017.3
blosc: None
bottleneck: None
tables: 3.4.2
numexpr: 2.6.4
feather: None
matplotlib: 2.1.2
openpyxl: None
xlrd: 1.1.0
xlwt: 1.3.0
xlsxwriter: None
lxml: None
bs4: 4.6.0
html5lib: 0.9999999
sqlalchemy: 1.2.2
pymysql: None
psycopg2: None
jinja2: 2.10
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: 0.6.0

[chris-b1]

For reasons not entirely clear to me, our xstrtod implementation can't parse the maximum 64 bit float. xref #17154, #19361

A workaround is to use the 'high precision' float parser - some discussion in #17154 about making that the default.

data = StringIO("""
a, b
1, 1
1.7976931348623157e+308, 1
""")
pd.read_csv(data, dtype={'a': float, 'b': float}, engine='c', float_precision='high')

[chris-b1]

Also applies to pd.to_numeric

pd.to_numeric(['1.7976931348623157e+308'])
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
pandas/_libs/src/inference.pyx in pandas._libs.lib.maybe_convert_numeric()

ValueError: Unable to parse string "1.7976931348623157e+308"

During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
<ipython-input-29-15b8074a875c> in <module>()
----> 1 pd.to_numeric(['1.7976931348623157e+308'])

~\AppData\Local\Continuum\Anaconda3\envs\py36\lib\site-packages\pandas\core\tools\numeric.py in to_numeric(arg, errors, downcast)
    131             coerce_numeric = False if errors in ('ignore', 'raise') else True
    132             values = lib.maybe_convert_numeric(values, set(),
--> 133                                                coerce_numeric=coerce_numeric)
    134 
    135     except Exception:

pandas/_libs/src/inference.pyx in pandas._libs.lib.maybe_convert_numeric()

ValueError: Unable to parse string "1.7976931348623157e+308" at position 0

[jreback]

this is not a bug, rather the exact reason for the high precision option. there is a similar issue IIRC. see if you can find it. cc @gfyoung

[gfyoung]

@jreback : Agreed. This is intended behavior.

@chris-b1 : However, your example with pd.to_numeric strikes me as interesting...I feel like we would want to support conversion to numeric with high-precision (if specified)? @jreback : Thoughts?

[jreback]

yeah i think to_numeric should use the high prevsion converters (though not sure if this goes thru the same path)

[gfyoung]

though not sure if this goes thru the same path

AFAICT, the xlstrtod functionality is just for the parsers.py file. Though if you want to integrate with it, you would probably have to add several more parameters to the to_numeric interface for it to provide similar functionality as in read_csv for example (e.g. decimal and sci as the floating point options used in tokenizer.h) OR perhaps we just hard-code values for those parameters.

[mroeschke]

Since the csv behavior is expected, repurposing this issue for the to_numeric enhancement