import numpy as np import pandas as pd # construct IntervalIndex from timestamps # --> Length of index is > 100, the default leaf_size # --> index is heavily skewed towards Timstamp.max idx = pd.IntervalIndex.from_arrays(pd.date_range('2018-01-01', '2018-12-31', freq='D'), [pd.Timestamp.max]*365) # type conversion from IntervalIndex._engine left = idx._maybe_convert_i8(idx.left) right = idx._maybe_convert_i8(idx.right) # pivot point calculation from IntervalNode constructor, when n_elements > leaf_size pivot = np.median(left + right) / 2 print(pd.Timestamp(pivot)) # >>> 1848-02-11 00:06:21.572612096 # This is below the minimum left-side value, which should not be print(pd.Timestamp(np.median(idx.mid.asi8))) # >>> 2140-05-21 23:53:38.427387904 # This is the correct midpoint # NB: with my own data, accessing `mid` throws OverflowError: Overflow in int64 addition # proposed alternate logic for midpoint alternate = np.median(left/2 + right/2) print(pd.Timestamp(alternate) # >>> 2140-05-21 23:53:38.427387904 # Matches the correct midpoint, and should never over/under flow

idx.right - idx.left # >>> Traceback (most recent call last): # >>> File "<input>", line 1, in <module> # >>> File "~\pandas\core\indexes\datetimelike.py", line 501, in __sub__ # >>> result = self._data.__sub__(maybe_unwrap_index(other)) # >>> File "~\pandas\core\arrays\datetimelike.py", line 1275, in __sub__ # >>> result = self._sub_datetime_arraylike(other) # >>> File "~\pandas\core\arrays\datetimes.py", line 724, in _sub_datetime_arraylike # >>> arr_mask=self._isnan) # >>> File "~\pandas\core\algorithms.py", line 938, in checked_add_with_arr # >>> raise OverflowError("Overflow in int64 addition") # >>> OverflowError: Overflow in int64 addition