Fix SklearnModelWrapper: get_feature_names() removed in sklearn 1.2+

[qiyanjun]

Summary

SklearnModelWrapper.__call__ calls self.tokenizer.get_feature_names() — but tokenizer here is actually a fitted CountVectorizer/TfidfVectorizer, and get_feature_names() was removed from scikit-learn in 1.2 (Dec 2022) in favor of get_feature_names_out() (available since 1.0). Any use of SklearnModelWrapper — including its own demo notebook, docs/2notebook/Example_1_sklearn.ipynb — has been broken with any scikit-learn released in the last several years, raising AttributeError on the first real prediction call.

Fix

• textattack/models/wrappers/sklearn_model_wrapper.py: get_feature_names() → get_feature_names_out().
• docs/2notebook/Example_1_sklearn.ipynb: same fix in the 4 places the notebook does its own vectorizer → DataFrame conversion. Also fixed !pip install datasets nltk sklearn → scikit-learn — the sklearn PyPI package name is deprecated and its installer now aborts with an explicit error directing users to scikit-learn instead.

The notebook fix was applied as a surgical text substitution (not a full cell replace), so the notebook's existing recorded outputs and execution counts — the demonstrated training accuracy / attack results from a real prior run — are left intact rather than wiped.

Verification

• Reproduced the AttributeError directly against the currently pinned scikit-learn (1.7.2): CountVectorizer().get_feature_names() → AttributeError: 'CountVectorizer' object has no attribute 'get_feature_names'
• Confirmed get_feature_names_out() works end-to-end through the actual wrapper (fit a real CountVectorizer + LogisticRegression, ran predictions through SklearnModelWrapper)
• Notebook JSON validated (json.load) after the edit; diff is minimal (5 lines) and doesn't touch stored outputs
• black --check / flake8 clean; from textattack.models.wrappers import SklearnModelWrapper still imports cleanly

🤖 Generated with Claude Code

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB