scikit-learn

Warnings

• breaking pip install sklearn now raises an error (since Dec 1, 2023). The sklearn PyPI package was deprecated and blocked. LLMs still generate pip install sklearn in requirements.txt and setup.py.
  Fix: Use pip install scikit-learn (with hyphen). Import remains: from sklearn.xxx import Yyy.
• breaking Pickled models are not guaranteed compatible across scikit-learn versions. Loading a model pickled with 0.21 into 1.x raises ValueError or produces silent wrong results.
  Fix: Always retrain models after upgrading scikit-learn. For long-lived model storage use ONNX or joblib with explicit version pinning and migration tests.
• breaking Python 3.10 support dropped in 1.8. scikit-learn 1.8 requires Python >=3.11.
  Fix: Pin scikit-learn<1.8 for Python 3.10 environments. Upgrade Python to 3.11+ for 1.8.
• breaking n_features_ attribute removed from most estimators. Replaced by n_features_in_ (added in 1.0). Code accessing clf.n_features_ raises AttributeError.
  Fix: Replace clf.n_features_ with clf.n_features_in_.
• gotcha Data leakage via manual transform: calling scaler.fit_transform(X_train) then scaler.transform(X_test) is correct, but calling scaler.fit_transform(X) before the split leaks test statistics into training. LLMs frequently generate this pattern.
  Fix: Always split first, then fit preprocessors only on X_train. Use Pipeline to enforce this automatically.
• gotcha random_state must be set on both train_test_split AND estimators for reproducible results. Setting only one still produces variation.
  Fix: Pass random_state=42 (or any fixed int) to: train_test_split, all estimators that accept it (RandomForestClassifier, LogisticRegression, etc.), and cross_val_score via cv parameter.
• gotcha pandas 3.0 string dtype compatibility: scikit-learn may fail or warn when receiving string-dtype columns (new default in pandas 3.0). Encode categoricals explicitly before fitting.
  Fix: Use OrdinalEncoder or OneHotEncoder within a Pipeline to handle string columns. Or convert: df['col'].astype('category') before passing to sklearn.

Install

• pip install scikit-learn Correct install name
• pip install sklearn WRONG — raises install error since Dec 2023

Imports

• sklearn

  from sklearn.linear_model import LogisticRegression
  from sklearn.model_selection import train_test_split
  from sklearn.preprocessing import StandardScaler

  Install with pip install scikit-learn (hyphen). Import with import sklearn (underscore). These are different strings — a permanent confusion source for LLMs.
• Pipeline

  from sklearn.pipeline import Pipeline
  
  pipe = Pipeline([
      ('scaler', StandardScaler()),
      ('clf', LogisticRegression())
  ])
  pipe.fit(X_train, y_train)

  Always use Pipeline to chain preprocessors and models. Manual fit/transform chains are error-prone: easy to accidentally fit on test data or forget to transform.

Quickstart

Correct pattern: Pipeline + train/test split. Always use random_state for reproducibility.

from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import Pipeline
from sklearn.metrics import accuracy_score

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

pipe = Pipeline([
    ('scaler', StandardScaler()),
    ('clf', LogisticRegression(random_state=42))
])
pipe.fit(X_train, y_train)

print(accuracy_score(y_test, pipe.predict(X_test)))