pad in arraypad.py changed behavior in an undocumented fashion #7353
Description
Previously this command worked:
a = [[1, 2], [3, 4]]
np.lib.pad(a, ((3, 2), (2, 3)), 'constant',
constant_values=((np.nan, np.nan), (np.nan, np.nan)))
However in the 1.10.4 it now returns the following stack trace:
Traceback (most recent call last):
File "C:/Users/Andrei/PycharmProjects/Pharmacosensitivity_growth_assays/src/post_processing.py", line 477, in <module>
all_cell_lines_arr, means_accumulator, errs_accumulator, names_accumulator = stack_data_in_range_of_interest(concs_effective_range)
File "C:/Users/Andrei/PycharmProjects/Pharmacosensitivity_growth_assays/src/post_processing.py", line 129, in stack_data_in_range_of_interest
mode='constant', constant_values=((np.nan, np.nan), (np.nan, np.nan)))
File "C:\Anaconda\lib\site-packages\numpy\lib\arraypad.py", line 1358, in pad
cast_to_int=False)
File "C:\Anaconda\lib\site-packages\numpy\lib\arraypad.py", line 1070, in _normalize_shape
return tuple(tuple(axis) for axis in arr.tolist())
AttributeError: 'tuple' object has no attribute 'tolist'
This is now the way of implementing the padding:
a = [[1, 2], [3, 4]]
nan_pad = np.array(((np.nan, np.nan), (np.nan, np.nan)))
np.lib.pad(a, ((3, 2), (2, 3)), 'constant',
constant_values=nan_pad)
Could you please add relevant documentation to the pad function in the arraypad.py?
An additional issue is that it looks like np.nan gets converted to -2147483648 which seems to be overflowing 32-bit signed integer. It does not affect my code that much, but this is an issue you might need to be aware of.