Orientation: How scikit-learn is put together

scikit-learn is organized around one shared contract: every estimator (classifier, regressor, transformer, or meta-estimator) implements fit/predict/transform and inherits parameter get/set behavior from a common base class. Algorithm modules (tree, linear_model, cluster, ...) plug into that contract, meta-estimators like Pipeline and GridSearchCV compose or wrap any conforming estimator, and performance-critical inner loops drop down into compiled Cython extensions. A small global config module threads cross-cutting behavior (validation, output format, metadata routing) through every fit call.