apply method, should return certain columns as datetime but returns them as int[64] instead

[jimbasquiat]

Code Sample, a copy-pastable example if possible

a=pd.DataFrame(data=list(range(10)))
def test(x):
    x["date"]=pd.Timestamp("2017-01-01")
    return x
a.apply(test,axis=1)

returns:

    0   date
0   0   1483228800000000000
1   1   1483228800000000000
2   2   1483228800000000000
3   3   1483228800000000000
4   4   1483228800000000000
5   5   1483228800000000000
6   6   1483228800000000000
7   7   1483228800000000000
8   8   1483228800000000000
9   9   1483228800000000000

Problem description

This should obviously return a dataframe with timestamps in the column date. Here it is having the numerical .value of the timestamp instead.

Please don't come up with answer like : " use df['date']=pd.Timestamp("2017-01-01"). "

I simplified a lot my problem as to focus on the datetime with .apply. I need to use .apply in my real life code.

Expected Output

    0   date
0   0   2017-01-01 00:00:00
1   1   2017-01-01 00:00:00
2   2   2017-01-01 00:00:00
3   3   2017-01-01 00:00:00
4   4   2017-01-01 00:00:00
5   5   2017-01-01 00:00:00
6   6   2017-01-01 00:00:00
7   7   2017-01-01 00:00:00
8   8   2017-01-01 00:00:00
9   9   2017-01-01 00:00:00

[jreback]

you are trying to mutate the internal Series. This violates all kinds of guarantes. Pandas cannot infer ever things a user will possible do. Not even sure what you are trying to accomplish.

In [7]: a=pd.DataFrame(data=list(range(10)))
   ...: def test(x):
   ...:     return pd.Timestamp("2017-01-01")
   ...: a.apply(test,axis=1)
   ...: 
Out[7]: 
0   2017-01-01
1   2017-01-01
2   2017-01-01
3   2017-01-01
4   2017-01-01
5   2017-01-01
6   2017-01-01
7   2017-01-01
8   2017-01-01
9   2017-01-01
dtype: datetime64[ns]

[jimbasquiat]

'you are trying to mutate the internal Series.' I think you are referring here to altering a list while iterating over it. this is def a different problem since the original dataframe is absolutely not modified by the .apply method.
'This violates all kinds of guarantes.' Could you point to documentation or information about that? i don't see any type inference related matters in the doc for .apply.
I understand this is an annoying burden to have to deal with this kind of requests for you, but then i would wonder why would you volunteer for taking charge of the issues log if its for dismissing the problems out of hand like that?

[jreback]

this is already reported in #15526

generally mutating things inside an apply, while it might work are not supported in any guaranteed way.

applying row-by-row is an anti-pattern when you can easily do a vectorized operation. Sure this may not be your real operation, but according to your SO post, your entire calculation can be vectorized.