Input validation: check_array and validate_data

check_array and check_X_y (composed together and wrapped by validate_data) form the shared input-validation contract that almost every estimator's fit/predict/transform routes through: they coerce dtypes, enforce 2D shape and minimum sample/feature counts, and reject NaN/Inf unless a config or parameter says otherwise. validate_data additionally records n_features_in_/feature_names_in_ on the estimator, so later calls can detect shape drift, while _fit_context wraps fit() to validate hyperparameters before this data validation ever runs. Understanding this pipeline explains where most of scikit-learn's input error messages come from and why every estimator behaves consistently on bad data.