v0.4.0

• Added eds_pseudo.dates_normalizer to parse ML detected dates and extract their value and format.
• Support empty doc._.context field
• Update EDS-NLP to v0.10.7:
  • fix somes issues with jsonl loading
  • more transformer overriding options
  • fix out-of-memory issues (auto split transformer input depending on the available memory)
  • fixes some multiprocessing deadlock issues
  • add chunk sorting option to the lazy collection set_processing method
• Replace gen_dataset/train.jsonl with the original fictitious templates and the dataset generation script.
• Update the README with the instructions to download the public pre-trained model.
• Improve packaging to add evaluation results to the model's meta field and packaged model README (for HF)

v0.2.0

Changelog

Many fixes along the publication of our article:

• committed the missing script infer.py
• changed config default bert model to camembert-base
• put config.cfg as a dependency, not params
• default to cpu training
• allow for missing metadata (i.e. omop's note_class_source_value)
• tests for the rule-based components
• code documentation and cleaning
• experiment and analysis scripts
• charts and tables in the Results page of our documentation

Pull Requests

• Article updates by @percevalw in #2
• chore: clean docs and scripts by @percevalw in #3
• docs: fix repository link by @bdura in #1
• fix: main branch CI + github badges by @percevalw in #4
• fix: add missing infer script and post-publication fixes by @percevalw in #6

New Contributors

• @percevalw made their first contribution in #2
• @bdura made their first contribution in #1

Full Changelog: https://github.com/aphp/eds-pseudo/commits/v0.2.0