Models

Type-directed model interfaces — the classifier / regressor typeclasses. Two Lean-style typeclasses give a *named, type-directed* interface over predictive models, so the same code drives a neural network and a scikit-learn model alike: * ``Classifier`` — a model whose target is a discrete class. ``classify`` returns a vector of per-class scores. * ``Regressor`` — a model whose target is a real number. ``regress`` returns the predicted number. The *concrete* model types are specific — ``NN.Network`` (a neural network), and ``Learning``'s ``SklearnClassifier`` / ``SklearnRegressor`` (fitted scikit-learn estimators) — and each opts into the matching class. The shared feature representation is a ``Tensor.Vector`` (one sample). Scope note: these generic method signatures do not specify dimension refinements. Consequently, the dimension invariants ``Tensor`` / ``NN`` and the shape/metric refinements ``Learning`` verify on their concrete operations are not carried as laws through these generic methods — a width mismatch surfaces at runtime, not compile time. Use the classes for type-directed dispatch and reuse; drop to ``NN.predict`` / ``Learning.predict`` directly for the static guarantee. ── Classes ──────────────────────────────────────────────────────────── The classifier typeclass: discrete target, vector of class scores.