Enhancing Machine Learning Workflows with skore
summary
skore presents a concrete ML workflow layer on top of scikitlearn and scrap, ships an estimator report that trains and stores models, a comparison reports object that compares metrics (baseline MAE → leaderboard 33.9 then improved to 31.6), and cross validation reports that run k‑fold. Demo shows permutation feature importance using two repeats and finds six dominant features, a ~20s cost for importance, and discussion of integrations with MLflow/ClearML and offloading heavy tasks to a hub.
AI-written summary · about
chapters (13)
- 0:00skore introduction
- 3:21What is skore
- 4:26Demo setup Kaggle
- 8:06transplit function warnings
- 9:10Baseline estimator report
- 11:21Table report data exploration
- 14:12Model comparison report
- 16:41Feature importance
- 21:44Cross-validation improvements
- 23:10skore artifacts purpose
- 24:02Q&A best practices
- 26:00Q&A experiment tracking
- 27:08Q&A computational efficiency
related talks
31mMLflow: Platform for Complete Machine Learning Lifecycle
57mManaging Machine Learning Lifecycle with MLflow
21mKeynote: Easier Machine Learning Thoughts From Scikit-Learn
34mBuilding End-to-End Machine Learning Workflows with Kubernetes...
15mKeynote: Machine Learning on Kubernetes Made Easy With Kubeflow
28mMachine Learning on Kubernetes at Shell: A Kubeflow Journey
40mMLSecOps with Automated Online and Offline ML Model Evaluations on Kubernetes
30mContainer workflows for data science and ML