ImportError after installing pandas==2.2.3 and numpy==1.26.4 on Google Colab (works after session restart)

[bernaljg]

I'm encountering a reproducible issue when installing specific versions of pandas and numpy in Colab.

My project depends on:

• pandas==2.2.3
• numpy==1.26.4.

These versions work perfectly in my local environment. However, in this Colab notebook, attempting to install and import them immediately results in an error when running import pandas.

What’s unusual is that if I restart the runtime and run all cells again, the error disappears and everything works as expected.

I'm running this in Chrome.

Here is the full stack trace for reference:

ValueError                                Traceback (most recent call last)
[<ipython-input-2-38d4b0363d82>](https://localhost:8080/#) in <cell line: 0>()
----> 1 import pandas

6 frames
[/usr/local/lib/python3.11/dist-packages/pandas/__init__.py](https://localhost:8080/#) in <module>
     35     ) from _err
     36 
---> 37 from pandas._config import (
     38     get_option,
     39     set_option,

[/usr/local/lib/python3.11/dist-packages/pandas/_config/__init__.py](https://localhost:8080/#) in <module>
     18     "warn_copy_on_write",
     19 ]
---> 20 from pandas._config import config
     21 from pandas._config import dates  # pyright: ignore[reportUnusedImport]  # noqa: F401
     22 from pandas._config.config import (

[/usr/local/lib/python3.11/dist-packages/pandas/_config/config.py](https://localhost:8080/#) in <module>
     66 import warnings
     67 
---> 68 from pandas._typing import (
     69     F,
     70     T,

[/usr/local/lib/python3.11/dist-packages/pandas/_typing.py](https://localhost:8080/#) in <module>
    196     int,
    197     np.ndarray,
--> 198     np.random.Generator,
    199     np.random.BitGenerator,
    200     np.random.RandomState,

[/usr/local/lib/python3.11/dist-packages/numpy/__init__.py](https://localhost:8080/#) in __getattr__(attr)
    335 
    336     def __dir__():
--> 337         public_symbols = globals().keys() | {'testing'}
    338         public_symbols -= {
    339             "core", "matrixlib",

[/usr/local/lib/python3.11/dist-packages/numpy/random/__init__.py](https://localhost:8080/#) in <module>
    178 
    179 # add these for module-freeze analysis (like PyInstaller)
--> 180 from . import _pickle
    181 from . import _common
    182 from . import _bounded_integers

[/usr/local/lib/python3.11/dist-packages/numpy/random/_pickle.py](https://localhost:8080/#) in <module>
----> 1 from .mtrand import RandomState
      2 from ._philox import Philox
      3 from ._pcg64 import PCG64, PCG64DXSM
      4 from ._sfc64 import SFC64
      5 

numpy/random/mtrand.pyx in init numpy.random.mtrand()

ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject```

[metrizable]

@bernaljg Thanks for filing the issue and thanks for using Colab.

With #5115, Colab is now pre-installed with NumPy 2. I note that your code does not initially work after installing an older version of NumPy, and this is because a more modern version is already loaded into the kernel to support variable inspection and quickcharting (numpy is a transitive dep of matplotlib):

matplotlib==3.10.0
├── contourpy [required: >=1.0.1, installed: 1.3.1]
│   └── numpy [required: >=1.23, installed: 2.0.2]
...

I also note that your install works as you expect after you restart your runtime, which then loads in the version of NumPy you've installed.

Although this may function, you may want to consider upgrading to a more modern version of NumPy, in light of its deprecation schedule. As well, in your case here, downgrading a core package like numpy may have unforeseen consequences for other packages and it leaves the runtime in an unsupported state for Colab.

[bernaljg]

I understand, thank you for the explanation!

We will work on upgrading to a more modern NumPy version.

[Asaf-Sivan]

Hello, i am having the same similar issue, is there any way to install numpy 1.26.4 without having to restart session? i want my colab to be as user friendly as possible and i am using packages that must use this numpy version that i don't have control on.