Improving production workflows for scikit-learn models with skops
summary
skops provides a reproducible repo layout, a zipped skops format that avoids pickle by serializing numpy/scipy/scikit‑learn objects, and a programmatic model card and config that record columns, examples and deps. loading requires an explicit trust flag or list of trusted types. push integrates with Hugging Face Hub to serve a hosted inference API and one‑line gradio demos. the serialization is experimental but usable now; inference API currently accepts pickle with skops support planned.
AI-written summary · about
chapters (12)
- 0:00SK Ops and model sharing
- 1:55Security and reproducibility
- 3:02Local repository with SK Ops
- 5:10Model configuration and cards
- 7:26Customizing model cards
- 9:56Secure serialization with SK Ops
- 12:07Loading models securely
- 12:48SK Ops with Hugging Face Hub
- 14:19Hosted inference API
- 16:26Gradual integration demos
- 19:16Resources and contributions
- 20:46Q&A
related talks
32mLet’s exploit pickle, and `skops` to the rescue!
28mEnhancing Machine Learning Workflows with skore
17mProductionizing Data Science Workloads
36mData Science Meets DevOps: Gitops with OpenShift
46mProduction Model Deployment
50mBuilding an MLOps strategy from the ground up
27mBuilding Secure MLOps Pipelines With KitOps + Argo Workflows
22mImprove Your ML Projects: Embrace Reproducibility and Production...