Why is predictproba unavailable for a hinge-loss SGDClassifier?
Classic MLLevel: IntermediateScore API compatibility
Why is predict_proba unavailable for a hinge-loss SGDClassifier?
Hints
- A real-valued margin need not lie in [0,1].
- Probability semantics require an additional model.
Answer
Hinge learning supplies margin scores, not a fitted probability link. Use decision_function for ranking or add leakage-safe calibration if probabilities are required.
Source and licence: scikit-learn 1.7.2 — sgd · BSD-3-Clause · answer: Quantoros editorial