ENH: json_normalize should allow a different separator than .

[jowens]

>>> import pandas
>>> col_in = ['c1', 'c2.x']
>>> df = pandas.DataFrame([['A', 0], ['B', 1]], columns=col_in)
>>> df.c1
0    A
1    B
Name: c1, dtype: object
>>> df.c2.x
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/pandas/core/generic.py", line 2744, in __getattr__
    return object.__getattribute__(self, name)
AttributeError: 'DataFrame' object has no attribute 'c2'

Problem description

The above snippet shows that it's not ideal to have . as a character in a column name. (I'm running into this when using Vega for data visualization, vega/vega-lite#1775.) When json_normalize flattens a nested input JSON, it separates the nesting levels with a .. I believe this happens on this line:

pandas/pandas/io/json.py

Line 831 in 7d8bc0d

meta_keys = ['.'.join(val) for val in meta]

I'd like to see an additional argument to json_normalize, separator, with default ., that specified the character (string) that separated nesting levels. In the line of code above, '.'.join(val) would be replaced by separator.join(val) (if I'm reading what that line does correctly). I could use, say, _ to use underscore instead of period.

n00b at pandas, please correct me if I'm doing anything wrong.

Expected Output

Output of pd.show_versions()

>>> pandas.show_versions()

INSTALLED VERSIONS

commit: None
python: 2.7.12.final.0
python-bits: 64
OS: Darwin
OS-release: 16.3.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: en_US.US-ASCII
LOCALE: None.None

pandas: 0.19.1
nose: 1.3.7
pip: 9.0.1
setuptools: 30.3.0
Cython: 0.25.2
numpy: 1.11.2
scipy: 0.18.1
statsmodels: None
xarray: None
IPython: 5.1.0
sphinx: 1.5.1
patsy: None
dateutil: 2.6.0
pytz: 2016.10
blosc: None
bottleneck: 1.2.0
tables: 3.2.3.1
numexpr: 2.6.1
matplotlib: 1.5.3
openpyxl: 2.4.1
xlrd: 1.0.0
xlwt: None
xlsxwriter: None
lxml: 3.6.0
bs4: 4.5.1
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.8
boto: None
pandas_datareader: None

[rcarneva]

The dot notation for accessing columns is just a convenience. You can still get the column normally using df['c2.x'].

[jowens]

I understand. I'm merely offering the observation that columns with . in the names are perhaps not a perfect fit for everything in Pandas and having the option to use a different separator might be also useful for someone who isn't me.

[rcarneva]

Sure thing. Just wanted to point that out in case it was keeping you from doing something you needed to do now since you said that you were new to pandas.

[jowens]

Yeah, it's the vega-lite bug I filed (vega/vega-lite#1775) that's my proximate difficulty here.

[jreback]

this would be quite easy to add; PR's welcome.