module 'numpy' has no attribute 'ndarray'

Question

My Jupiter notebook was crushed, so I have to reinstall the notebook, but in the new Jupiter notebook, I cannot run pandas.

import pandas as pd

AttributeError                            Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_6860/4080736814.py in <module>
----> 1 import pandas as pd

~\anaconda3\lib\site-packages\pandas\__init__.py in <module>
     20 
     21 # numpy compat
---> 22 from pandas.compat import (
     23     np_version_under1p18 as _np_version_under1p18,
     24     is_numpy_dev as _is_numpy_dev,

~\anaconda3\lib\site-packages\pandas\compat\__init__.py in <module>
     12 import warnings
     13 
---> 14 from pandas._typing import F
     15 from pandas.compat.numpy import (
     16     is_numpy_dev,

~\anaconda3\lib\site-packages\pandas\_typing.py in <module>
     82 # array-like
     83 
---> 84 ArrayLike = Union["ExtensionArray", np.ndarray]
     85 AnyArrayLike = Union[ArrayLike, "Index", "Series"]
     86 

AttributeError: module 'numpy' has no attribute 'ndarray'

I have tried to rename or delete the numpy.py, but it didn't work.

Answer 1

The problem is not with Pandas. This is due to NumPy. I had a similar issue and this is what I did.

Ran python -c "import numpy as np; print(np.__file__); print(np.ndarray)". The expect output will contain the location of your installed NumPy package and <class 'numpy.ndarray'> for the second print statement. In my case, it seemed that NumPy package was located in ~/.local.

I tried reinstalling NumPy in conda, but the location was still the same. So, as per this solution, I deleted ~/.local. Then the error disappeared.

Answer 2

For me none of the above solutions fixed the issue. The only way it worked was I cleaned up all my python environments.

• Removed manually installed python versions at various locations such as usr/local/bin
• cleaned up PATH
• Removed all but python 3.11 from brew.
• Ran brew cleanup and resolved all warnings/issues.
• Re-installed command line tools since I had been using the system installed python version for various projects.
• Created a fresh virtual environment using brew's python3.11 as the base interpreter. Post this both numpy and pandas were not present and were installed fresh. No issues anymore.