sap-ai-research

/

contexttab

like 3

Follow

SAP AI Research 71

@MaximilianSchambach , @nielsr : Just wanted to give you a heads-up on running your Kaggle and Colab notebooks.

I'm encountering issues with package installations and imports, which are preventing the notebook from running correctly.

1. Dependency Conflicts (from pip):
First, pip is reporting numerous dependency resolution errors. Key conflicts include:

• torch ecosystem: My installed torch 2.7.1 is incompatible with torchaudio, torchvision, and fastai, all of which expect torch 2.6.0 (or an earlier version).
• pyarrow: My pyarrow 20.0.0 is too new for cudf/pylibcudf (which need an older version).
• scikit-learn: My scikit-learn 1.7.0 is too new for category-encoders and sklearn-compat.
• google-colab related packages: There are conflicts with google-colab's expected versions of pandas, google-auth, requests, tornado, and notebook.
• numpy: My numpy 1.26.4 is incompatible with cesium (which expects numpy 2.x).

2. torchvision Circular Import Error:
Following the installation attempts, I'm also getting an AttributeError: partially initialized module 'torchvision' has no attribute 'extension' (most likely due to a circular import).

Let me know your thoughts on how best to proceed!

Colab:
AttributeError: partially initialized module 'torchvision' has no attribute 'extension' (most likely due to a circular import)
/usr/local/lib/python3.11/dist-packages/torchvision/_meta_registrations.py in wrapper(fn)
16 def register_meta(op_name, overload_name="default"):
17 def wrapper(fn):
---> 18 if torchvision.extension._has_ops():
19 get_meta_lib().impl(getattr(getattr(torch.ops.torchvision, op_name), overload_name), fn)
20 return fn

Kaggle:
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
bigframes 2.8.0 requires google-cloud-bigquery-storage<3.0.0,>=2.30.0, which is not installed.
pylibcudf-cu12 25.2.2 requires pyarrow<20.0.0a0,>=14.0.0; platform_machine == "x86_64", but you have pyarrow 20.0.0 which is incompatible.
cudf-cu12 25.2.2 requires pyarrow<20.0.0a0,>=14.0.0; platform_machine == "x86_64", but you have pyarrow 20.0.0 which is incompatible.
datasets 3.6.0 requires fsspec[http]<=2025.3.0,>=2023.1.0, but you have fsspec 2025.5.1 which is incompatible.
category-encoders 2.7.0 requires scikit-learn<1.6.0,>=1.0.0, but you have scikit-learn 1.7.0 which is incompatible.
cesium 0.12.4 requires numpy<3.0,>=2.0, but you have numpy 1.26.4 which is incompatible.
google-colab 1.0.0 requires google-auth==2.38.0, but you have google-auth 2.40.3 which is incompatible.
google-colab 1.0.0 requires notebook==6.5.7, but you have notebook 6.5.4 which is incompatible.
google-colab 1.0.0 requires pandas==2.2.2, but you have pandas 2.2.3 which is incompatible.
google-colab 1.0.0 requires requests==2.32.3, but you have requests 2.32.4 which is incompatible.
google-colab 1.0.0 requires tornado==6.4.2, but you have tornado 6.5.1 which is incompatible.
sklearn-compat 0.1.3 requires scikit-learn<1.7,>=1.2, but you have scikit-learn 1.7.0 which is incompatible.
pandas-gbq 0.29.1 requires google-api-core<3.0.0,>=2.10.2, but you have google-api-core 1.34.1 which is incompatible.
torchaudio 2.6.0+cu124 requires torch==2.6.0, but you have torch 2.7.1 which is incompatible.
torchvision 0.21.0+cu124 requires torch==2.6.0, but you have torch 2.7.1 which is incompatible.
fastai 2.7.19 requires torch<2.7,>=1.10, but you have torch 2.7.1 which is incompatible.
bigframes 2.8.0 requires google-cloud-bigquery[bqstorage,pandas]>=3.31.0, but you have google-cloud-bigquery 3.25.0 which is incompatible.
bigframes 2.8.0 requires rich<14,>=12.4.4, but you have rich 14.0.0 which is incompatible.
jupyter-kernel-gateway 2.5.2 requires jupyter-client<8.0,>=5.2.0, but you have jupyter-client 8.6.3 which is incompatible.

Fix:
Applying the following resolves the issue. Could you please update the Colab?

!pip uninstall torch torchvision -y
!pip install torch==2.1.0 torchvision==0.16.0 --index-url https://download.pytorch.org/whl/cu121
!pip install torchaudio

Classification Task:
Start async server to compute embedding on port 5655
Warning: no response from server (Resource temporarily unavailable).
You might have forgotten to start it, or it might have been killed?
Trying restarting it.
Start async server to compute embedding on port 5655
Accuracy 0.9789473684210527

Regression Task:
Start async server to compute embedding on port 5655
Port already in use, not starting again.
R² Score: 0.8927975982420193

Note: Please check for the correct dependencies.