scikit-learn and Pandas preview

scikit-learn estimators accept NumPy arrays; Pandas DataFrames work when all columns are numeric. Export with to_numpy(), keep feature names in ColumnTransformer pipelines, and never leak test statistics into train features.

Feature matrix convention

import pandas as pd
import numpy as np

df = pd.DataFrame({'age': [25, 30], 'income': [50000, 60000]})
X = df.to_numpy()  # shape (n_samples, n_features)
print('X shape:', X.shape)

Pipeline pattern

• Split train/test before fit
• Fit scalers/encoders on train only
• Use ColumnTransformer for mixed numeric/categorical columns
• Keep column names in a list parallel to X columns

Common pitfall

Fitting StandardScaler on full dataset before split leaks future information—always fit on train, transform both.

Important interview questions and answers

1. Q: X shape?
   A: (n_samples, n_features)—rows are observations, columns are features.
2. Q: Categorical columns?
   A: One-hot encode or ordinal encode before sklearn—does not accept raw strings.

Self-check

1. Export a numeric DataFrame to X matrix.
2. Why fit scaler on train only?

Pitfall: Fit scalers and encoders on train split only—never on full data before split.