Series with dtype=object does unexpected type conversion

[AllenDowney]

Code Sample, a copy-pastable example if possible

# Example 1

timestamp = pd.Timestamp(1412526600000000000)
series = pd.Series([], dtype=object)
series['timestamp'] = timestamp
type(series.timestamp)

# Example 2

series = pd.Series([], dtype=object)
series['anything'] = 300.0
series['timestamp'] = timestamp
type(series.timestamp)

Problem description

In the first example, the timestamp is still a Timestamp.

In the second example, the timestamp gets converted to int.

Expected Output

I expected the timestamp to continue to be a Timestamp, especially because the dtype of the Series is object. Why are the types of the values getting converted?

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.6.5.final.0
python-bits: 64
OS: Linux
OS-release: 4.4.0-128-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.utf8
LOCALE: en_US.UTF-8

pandas: 0.23.0
pytest: 3.5.1
pip: 10.0.1
setuptools: 39.1.0
Cython: 0.28.2
numpy: 1.14.3
scipy: 1.1.0
pyarrow: None
xarray: None
IPython: 6.4.0
sphinx: 1.7.4
patsy: 0.5.0
dateutil: 2.7.3
pytz: 2018.4
blosc: None
bottleneck: 1.2.1
tables: 3.4.3
numexpr: 2.6.5
feather: None
matplotlib: 2.2.2
openpyxl: 2.5.3
xlrd: 1.1.0
xlwt: 1.3.0
xlsxwriter: 1.0.4
lxml: 4.2.1
bs4: 4.6.0
html5lib: 1.0.1
sqlalchemy: 1.2.7
pymysql: None
psycopg2: None
jinja2: 2.10
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None

[WillAyd]

This is probably the same issue as #13910

[AllenDowney]

Thanks for the quick reply! Yes, looks like it is the same issue. Any suggestions on a workaround? I thought with `dtype=object`, Series might stop trying to infer types. Is there any way to just turn off type inference?

…

On Thu, Jul 12, 2018 at 5:16 PM, William Ayd ***@***.***> wrote: This is probably the same issue as #13910 <#13910> — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#21881 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABy37bcl-Bm3NSA-rhUPCfMDlneKNNgUks5uF7y7gaJpZM4VNvfU> .

[makbigc]

When adding a new index, np.concatenate joins the new array of value to the old one in the function _setitem_with_indexer.

np.concatenate turns the array with dtype='datetime64[ns] into int. While it doesn't turn the array with dtype=object.

Explicitly,

In [132]: array([pd.Timestamp('2016-01-01')])
Out[132]: array([Timestamp('2016-01-01 00:00:00')], dtype=object)

In [133]: np.concatenate([array([None]), array([pd.Timestamp('2016-01-01')])])
Out[133]: array([None, Timestamp('2016-01-01 00:00:00')], dtype=object)

In [134]: a = Series(pd.Timestamp('2016-01-01'))._values

In [135]: a
Out[135]: array(['2016-01-01T00:00:00.000000000'], dtype='datetime64[ns]')

In [136]: np.concatenate([array([None]), a])
Out[136]: array([None, 1451606400000000000], dtype=object)

Should a type check be added before np.concatenate? But it makes the code redundant.

[mroeschke]

This looks to work on master now. Could use a test

In [81]: timestamp = pd.Timestamp(1412526600000000000)
    ...: series = pd.Series([], dtype=object)
    ...: series['timestamp'] = timestamp
    ...: type(series.timestamp)
Out[81]: pandas._libs.tslibs.timestamps.Timestamp

In [82]: series = pd.Series([], dtype=object)
    ...: series['anything'] = 300.0
    ...: series['timestamp'] = timestamp
    ...: type(series.timestamp)
Out[82]: pandas._libs.tslibs.timestamps.Timestamp

[arongergely]

take