There is an inconsistency between how Series and DataFrames export types to dicts. The issue only manifests if the dataframe has multiple columns with different types. I think the issue is probably pretty far under the surface with differences between Series and DataFrame dtypes and just happened to be uncovered by export to dict.

Code Sample, a copy-pastable example if possible

import pandas as pd orig_dict = [{0: 'a', 1: 1}] # does not happen with # orig_dict =[{0: 0, 1: 1}] print 'Original dict' for k in orig_dict[0]: print type(orig_dict[0][k]) print df = pd.DataFrame(orig_dict) dict_exported_from_dataframe = [df.loc[[x]].to_dict(orient='records')[0] for x in df.index] dict_exported_from_series = [df.loc[x].to_dict() for x in df.index] print 'Dict exported from DataFrame' for k in dict_exported_from_dataframe[0]: print type(dict_exported_from_dataframe[0][k]) print print 'Dict exported from Series' for k in dict_exported_from_series[0]: print type(dict_exported_from_series[0][k]) print 

Expected Output

Original dict <type 'str'> <type 'int'> Dict exported from DataFrame <type 'str'> <type 'int'> Dict exported from Series <type 'str'> <type 'int'> 

OR

Original dict <type 'str'> <type 'int'> Dict exported from DataFrame <type 'str'> <type 'numpy.int64'> Dict exported from Series <type 'str'> <type 'numpy.int64'> 

output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 2.7.10.final.0
python-bits: 64
OS: Darwin
OS-release: 14.5.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8

pandas: 0.18.1
nose: None
pip: 8.1.2
setuptools: 25.2.0
Cython: None
numpy: 1.11.1
scipy: 0.13.0b1
statsmodels: None
xarray: None
IPython: 5.1.0
sphinx: None
patsy: None
dateutil: 2.5.3
pytz: 2016.6.1
blosc: None
bottleneck: None
tables: None
numexpr: None
matplotlib: None
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: None
boto: None
pandas_datareader: None