medcat/v2.0.0

We’re excited to announce the release of MedCAT v2. This is a major refactor that brings a more modular, flexible, and maintainable foundation for clinical NLP, while staying compatible with existing v1 models.

This release focuses on:

• Refactored structure for lower coupling and greater extensibility
• Modularity via optional install extras (install only what you need)
• Improved flexibility in tokenization, NER, and annotation pipelines
• Backwards compatibility for v1 models, with automatic conversion

✨ What’s New

• Decoupled from spacy → now possible to use lightweight regex tokenizer or other (custom) backends
• Optional extras: install support only for the components you need (spacy, meta-cat, deid, rel-cat, dict-ner)
• Training is now structured around dedicated classes for clearer workflows
• Tutorials and scripts have been rebuilt from the ground up for v2
• Added support for a supervised training web service (experimental, under development)

⚠️ Breaking Changes

• Saving/Loading:
  • Save method has a new name (CAT.save_model_pack)
  • v2 saves models in a new format (but still loads v1 models, with slower load times due to conversion)
• Training:
  • Training APIs now go through separate trainer classes
• Defaults:
  • Default install no longer includes spacy or advanced components (see migration guide for how to enable them)

For a complete list, see: BREAKING_CHANGES.md

📖 Migration Guide

If you’re upgrading from v1, please read the dedicated Migration Guide. It covers:

• Installation instructions
• Changes to saving/loading
• v1 model compatibility notes
• Updated tutorials and example scripts
• FAQ and troubleshooting

🔗 Useful Links

📦 PyPI

📘 Documentation

🧩 Migration Guide

💬 Discussion Forum

🛠️ Repository

Feedback

v2 is a big step forward, and we’d love your input!
Please open a GitHub issue or join the discussion forum for:

• Missing documentation
• Bugs or breaking behaviour
• Feedback on error/log messages
• Suggestions for future improvements