Open
Description
mean() works on timedelta64, but gives an inadvertent error message for datetime64:
In [16]: x = np.array(['2000-01-01', '2000-01-10'], dtype='datetime64[us]')
In [17]: np.mean(x)
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-17-67ce1f9814a2> in <module>
----> 1 np.mean(x)
~/miniconda3/envs/numbagg-py37/lib/python3.7/site-packages/numpy/core/fromnumeric.py in mean(a, axis, dtype, out, keepdims)
3116
3117 return _methods._mean(a, axis=axis, dtype=dtype,
-> 3118 out=out, **kwargs)
3119
3120
~/miniconda3/envs/numbagg-py37/lib/python3.7/site-packages/numpy/core/_methods.py in _mean(a, axis, dtype, out, keepdims)
73 is_float16_result = True
74
---> 75 ret = umr_sum(arr, axis, dtype, out, keepdims)
76 if isinstance(ret, mu.ndarray):
77 ret = um.true_divide(
TypeError: ufunc add cannot use operands with types dtype('<M8[us]') and dtype('<M8[us]')
The problem is that mean is implemented via sum, which cannot be sensibly defined for dates.
Fixing this will probably require some significant refactoring -- we would need to make mean something that could be overwritten explicitly based on the dtypes. Our usual mechanism for this is ufuncs, but mean is currently written in terms of ufuncs (add.reduce and divide) rather being a ufunc itself.