Probabl's open-source skore library makes ML evaluation and experiment tracking a default workflow

GaelVaroquaux · x · 2026-09-16

Probabl promotes skore, an open-source library for data scientists using scikit-learn and other ML frameworks, turning methodological best practices into workflow defaults.

Its getting-started guide walks through a full binary classification workflow (toxicity dataset): proper train/test setup, cross-validated development and comparison of multiple models, final validation on held-out data, and experiment tracking with smart diagnostics. Reports can be pushed to Skore Hub for storing, sharing and exploring.

Original post →

More from coding & agent

coding & agent channel →