An update on the latest scikit-learn features
summary
Changelog-level details: scikit-learn 1.4–1.5 adds an metadata routing API that lets top-level tools propagate sample_weight and groups into nested GridSearchCV/CV when enabled (experimental flag), a TunedThresholdClassifierCV that fits a classifier and then optimizes decision thresholds against a custom utility scorer that can consume metadata like transaction amounts, better native handling of categorical and missing values in histogram boosters, and provisional array API support that yields a ~40s→440ms GPU speedup for some estimators but remains optional and limited in coverage.
AI-written summary · about
chapters (8)
- 0:00scikit-learn latest features
- 0:44Metadata routing API
- 11:52Metadata for decisions
- 18:17Data types and hardware
- 21:52Categorical and missing data
- 24:15Q&A GPU support
- 27:11Fairness metrics and routing
- 29:19Roadmap and priorities
related talks
28mGet the best from your scikit-learn classifier
35mScikit-Learn can do THAT?!
21mKeynote: Easier Machine Learning Thoughts From Scikit-Learn
39mPushing Cython to its Limits in Scikit-learn
32mDeep Dive into scikit-learn's HistGradientBoosting Classifier..
48mOptimize the Right Thing
35mPandas, Pipelines, and Custom Transformers
36mScikit-learn on GPUs with Array API