1.16. Probability calibration
A probability-calibration guide covering reliability curves and sigmoid, isotonic and temperature-scaling approaches.
External resource. No execution or independent verification is claimed here.
How do calibration methods trade off log loss and reliability when the calibration sample is small?
An evaluator holds out calibration and test sets separately, recomputes probability metrics and retains the chosen binning and uncertainty estimates.
What you could produce
- A versioned minimal protocol, evaluator-owned test cases, and a comparison report with numerical/statistical uncertainty and failures.
Before you use it
- scikit-learn 1.9.1
- Compatible NumPy/SciPy and compiled dependencies; exact environment not prepared
Limits to keep in view
- No upstream example, source package, build hook or submitted code was executed.
- The current qualified pilot accepts only self-contained Python 3.13 with a 90-second deadline. This reference's package/runtime is not qualified for that path.
- The proposed protocol requires bounded resource estimates and an independently controlled evaluator before any scientific execution claim.
Source and permission context
scikit-learn documentation. 1.16. Probability calibration. https://scikit-learn.org/stable/modules/calibration.html; observed version 1.9.1 (observed documentation version; stable URL is mutable).
Rights need review. Review the scope and upstream conditions before reuse.
documentation · BSD-3-Clause
This documentation page credits the developers and explicitly labels its footer BSD License; the matching 1.9.1 COPYING file identifies the three-clause terms. Referenced datasets and separately licensed material are not cleared.
Inspect the license evidence ↗Still unresolved
- Third-party figures, linked papers, datasets and dependency licenses have not been assessed; no external content is copied into this catalog.