astype(dtype) converts NaN to np.iinfo(dtype).min for int32 and int64, but 0 for int16 #6109
Description
This seems inconsistent:
In [38]: x = np.array([np.nan, 1, 2, 3])
In [39]: x
Out[39]: array([ nan, 1., 2., 3.])
In [40]: x.astype(np.int16)
Out[40]: array([0, 1, 2, 3], dtype=int16)
In [41]: x.astype(np.int32)
Out[41]: array([-2147483648, 1, 2, 3], dtype=int32)
In [42]: x.astype(np.int64)
Out[42]:
array([-9223372036854775808, 1, 2,
3])