DataExcept replaces the generic ValueError and RuntimeError that data pipelines usually raise with a hierarchy of specific, catchable exception types: data loading and validation, missing columns, feature engineering, model training and convergence, inference, and the other operational boundaries of a data science or machine learning system. Each exception carries context (the field, the model, the dataset), exports to JSON against a published, versioned schema so non-Python consumers can validate it, and pickles cleanly so it survives a process boundary.
Install
1
pip install DataExcept
Project Links
- PyPI: DataExcept
- Documentation: diogoribeiro7.github.io/DataExcept
- Source: github.com/DiogoRibeiro7/DataExcept
- Issues: github.com/DiogoRibeiro7/DataExcept/issues
- Changelog: GitHub releases
Package Metadata
- Current release:
1.6.0 - Requires Python:
>=3.10,<3.15
Where It Fits
Use it in pipelines and libraries where a caller needs to distinguish "the input was wrong" from "the model did not converge" from "inference ran out of memory" without parsing message strings. It is the exception layer that industrialstats and other packages on this site standardise on, and it is a good fit for services that log or forward failures to systems written in other languages.