Fixed SMoG examples

[thomasmarchioro3]

This PR addresses some minor fixes in the SMoG example scripts and notebook (examples/notebooks/pytorch/smog.ipynb, examples/notebooks/pytorch_lightning/smog.ipynb, examples/pytorch/smog.py, examples/pytorch_lightning/smog.py):

OS and enviroment:

• Ubuntu 24.04.2 LTS
• Python 3.12.7
• lightly==1.5.22
• scikit-learn==1.7.1
• torch==2.7.1
• pytorch-lightning==2.5.2

---

Fix 1: Missing scikit-learn requirement

• Issue: scikit-learn was not listed in the example dependencies, leading to ImportError when running the code.
• Fix: Added scikit-learn>=1.7.1 to the example installation instructions.
  Note: Consider also adding this to the default dependencies in pyproject.toml.

---

Fix 2: KMeans ValueError due to incorrect tensor shape

• Issue: The KMeans call fails with:

kmeans = KMeans(self.n_groups).fit(features) raises
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 ValueError: n_samples=128 should be >= n_clusters=300.

caused by an incorrect array shape — features were shaped (num_features, memory_size) instead of (memory_size, num_features).

• Fix: Corrected array shape by transposing the feature array:

features = features.cpu().numpy().T

---

Fix 3: Incorrect import in pytorch_lightning examples

• Issue: Outdated MemoryBankModule + missing feature shape

self.memory_bank = loss.memory_bank.MemoryBankModule(size=memory_bank_size)

• Fix: Used correct module (same as pytorch examples)

---

Additional minor fixes

• Added #type: ignore when converting features from tensor to numpy array to avoid warnings from the language server

[liopeer]

Hey @thomasmarchioro3 and thank you for the contribution! We will have a look at your PR in the next few days :)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (master@1d2eb09). Learn more about missing BASE report.
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff            @@
##             master    #1876   +/-   ##
=========================================
  Coverage          ?   86.07%           
=========================================
  Files             ?      167           
  Lines             ?     6966           
  Branches          ?        0           
=========================================
  Hits              ?     5996           
  Misses            ?      970           
  Partials          ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[yutong-xiang-97]

Hi @thomasmarchioro3, thank you for your contribution!

Quick note: in example/ the notebooks are auto-generated from the example scripts for pytorch, pytorch-lightning, and pytorch-lightning-distributed. So whenever you add or modify examples, make sure that:

1. you add or update the script examples for all three of them, then
2. generate corresponding notebooks with make generate-example-notebooks following our contributing guide: https://github.com/lightly-ai/lightly/blob/master/CONTRIBUTING.md

Then the "check example notebooks" test will pass.

[yutong-xiang-97]

Great catch in finding the dimension mistake!

For scikit-learn dependency, your solution is sufficient since SMoG is not officially added to LightlySSL yet, and it is not used anywhere else.

[thomasmarchioro3]

Thanks for the suggestions! I hadn't notice that the feature tensor was being transposed in _reset_group_features.

I also have removed the # type: ignore. It causes some warnings from LSPs (like Pylance or Pyright) but that's not an actual issue, since examples are not type checked.

[yutong-xiang-97]

Approved!