v0.6.5

• Renard now uses uv for dependency managements
• Dropped support for Python 3.8, added support for Python 3.12

v0.6.3

• Update transformers to avoid non-relevant warning messages (huggingface/transformers#29358)
• Stop explicitly supporting stanza and spacy extras to simplify dependency management

v0.6.2

• CoOccurrencesGraphExtractor now keeps the type of the entity as a node attribute entity_type. This is useful when extracting a graph with mixed entity types.
• The dynamic Gephi export now supports extracting existing dynamic node attributes, including entity_type.

v0.6.1

• Graph plotting
  • Add tight_layout argument to plot functions
  • Add legend argument to plot functions
• Context retrieval
  • Add several context retrievers for BERTNamedEntityRecognizer
    • NERNeighborsContextRetriever
    • NERBM25ContextRetriever
    • NERNeuralContextRetriever
  • Fix wrong NERContextRetriever context mask
• GraphRulesCharacterUnifier
  • Improve coreference resolution linking rule
  • Add ignore_leading_determiner option
• Add determiners resources
• Add max_sent_len parameter to load_conll2002_bio
• Update to Tibert v0.5. Generally update dependency constraints to be more lenient
• Support for Python 3.11

v0.5.0

• BertForCoreferenceResolution : update to Tibert v0.4.0 and support for hierarchical merging.
• Pipeline progress report is now cleaner less overwhelming.
• Improved error messages when a step is missing a dependency.
• Graph plotting:
  • Graph plotting is now fully configurable by passing arguments to networkx plotting functions.
  • Tweaks to default plotting parameters.
• BlockBounds has been added to support custom blocks in general.
  • It is now possible to specify a custom dynamic window by supplying the dynamic_blocks argument!
  • Similary, CoOccurrencesGraphExtractor now supports specifying custom co-occurrence windows with the co_occurrences_blocks argument.
  • See the documentation and the new renard.utils.block_bounds function for more details.
• ConversationalGraphExtractor can now extract directed mention networks!
• Steps can now modify some pipeline parameters at init time. This allows steps to configure the global behavior of the pipeline. For example, depending on the model loaded, BertNamedEntityRecognizer can set the pipeline character_ner_tag to inform the next steps of the NER tag corresponding to persons entity.
• Add a test for the Stanford CoreNLP pipeline.
• Bugfixs:
  • Fix a rare crash in GraphRulesCharacterUnifier.
  • Fix a rare crash in NERDataset.
  • All warnings are now correctly printed to stderr.
• Documentation update.

v0.4.2

• Add the ignore_lone_titles parameter to GraphRulesCharacterUnifier

v0.4.1

• Mention equality is now correctly computed
• Fix a crash in Renard's tutorial
• Documentation improvements

v0.4.0

• Add unsupervised context retriever for BertNamedEntityRecognizer
• Greatly improve GraphRulesCharacterUnifier performance
• Update documentation, with a new "Contributing" section
• Add some utilities to easily train a custom NER model
• Add a new french NER model! compnet-renard/bert-base-cased-literary-NER
• The characters_graph state attribute has been deprecated (use character_network instead)
• Add py.typed file to better support typing in external tools
• Update tibert to version 0.3.0
• Minor fixes

v0.3.1

• fix several typos in the code and documentation
• BertNamedEntityRecognizer now correctly reports progress
• improve documentation
• security fixs
• fix a bug where the chapters argument would cause Renard to crash

v0.3.0

• Allow passing custom model and tokenizer for BERT based steps
• Add the ability to extract conversational networks thanks to three new steps (experimental)
• Custom steps can now return custom attributes
• Huge enhancements for GraphRulesCharactersExtractor
• A few enhancements for plotting : one can now always pass custom layouts, 0 degree nodes are now correctly displayed, better defaults
• A lot of minor fixs and QOL changes

Pypi release : https://pypi.org/project/renard-pipeline/0.3.0/