Possible bug in df.update

[isaacgg]

Code Sample, a copy-pastable example if possible

test_json = [{"_id": 'a', 'date': datetime.now()}, {"_id": 'b', 'date': datetime.now()}]
test_df = pd.DataFrame(test_json)

new_df = test_df.copy()
new_df["date"] = None
new_df.update(test_df)

print(test_df.head())
print(new_df.head())

Problem description

When using update function with datetime data, it is automatically converted to timestamp, which for me it seems like an abnormal behaviour. Code from above would output

_id date
0 a 2019-11-07 15:50:06.072158
1 b 2019-11-07 15:50:06.072158
_id date
0 a 1573141806072158000
1 b 1573141806072158000

Expected Output

_id date
0 a 2019-11-07 15:50:06.072158
1 b 2019-11-07 15:50:06.072158
_id date
0 a 2019-11-07 15:50:06.072158
1 b 2019-11-07 15:50:06.072158

Output of pd.show_versions()

[paste the output of pd.show_versions() here below this line]
INSTALLED VERSIONS

commit : None
python : 3.7.4.final.0
python-bits : 64
OS : Windows
OS-release : 10
machine : AMD64
processor : Intel64 Family 6 Model 61 Stepping 4, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.None

pandas : 0.25.1
numpy : 1.16.4
pytz : 2019.2
dateutil : 2.8.0
pip : 19.2.2
setuptools : 41.0.1
Cython : 0.29.13
pytest : None
hypothesis : None
sphinx : 2.1.2
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.4.1
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.10.1
IPython : 7.7.0
pandas_datareader: None
bs4 : None
bottleneck : None
fastparquet : None
gcsfs : None
lxml.etree : 4.4.1
matplotlib : 3.1.1
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pytables : None
s3fs : None
scipy : 1.3.1
sqlalchemy : None
tables : None
xarray : None
xlrd : None
xlwt : None
xlsxwriter : None

[susan-shu-c]

Hi, I was able to reproduce this result. This is due to pandas.DataFrame.update calling expressions.where. source link.

From then it eventually calls numpy.where documentation which then eventually uses the Numpy MaskedArray type. source link.

This seems to be a choice to use numpy.where, which causes the datetime type to be converted to unix time, which speeds up the computation. However feel free to correct me on that.

I'd suggest trying pandas.to_datetime linked here to convert them afterward (sometimes you have to reduce your unix time precision, by removing digits from the end, to get it to work), but I haven't tested on your example data yet, so feel free to try it.

[rhshadrach]

Indeed, this appears to be an odd interaction with DatetimeArray and np.where.

a = np.asarray([None], dtype=np.object)
b = np.asarray(pd.arrays.DatetimeArray(pd.Series([datetime.now()])))
cond = [False]
print(np.where(cond, a, b))

gives [1595782471507905000]; whereas

a = np.asarray([None], dtype=np.object)
b = np.asarray([datetime.now()], dtype=np.object)
cond = [False]
print(np.where(cond, a, b))

gives [datetime.datetime(2020, 7, 26, 16, 53, 4, 806281)]