How franck-dernoncourt/neuroner works
NeuroNER is a neural named-entity recognition application. Users provide CoNLL or BRAT datasets, then train a model or apply a pretrained model to test or deployment text; predictions can be evaluated and converted for BRAT use. The graph includes documented data/model fetching and format-conversion capabilities. Some orchestration and conversion wiring is inferred from documented behavior because the relevant source excerpts are partial.
Main entry points: neuromodel.py NeuroNER API · train.py Training and prediction · __main__.py CLI · dataset.py Dataset loader
Components 15
Entry and orchestration
- CLI
neuroner/__main__.py - NeuroNER API
neuroner/neuromodel.py
Dataset preparation
- Dataset loader
neuroner/dataset.py - BRAT to CoNLL
neuroner/brat_to_conll.py - CoNLL to BRAT
neuroner/conll_to_brat.py - i2b2 XML converter
neuroner/data/i2b2_2014_deid/xml_to_brat.py
NER modeling
- Entity LSTM
neuroner/entity_lstm.py - Training and prediction
neuroner/train.py
Evaluation and artifacts
- Pretrained models
neuroner/trained_models - Metrics and plots
neuroner/evaluate.py - Model sharing prep
neuroner/prepare_pretrained_model.py - Prediction outputs
Other components
- User
- Tokenizer services
- Pretrained embeddings
Connections 16
- User → CLI: runs
- User → NeuroNER API: calls API
- CLI → NeuroNER API: dispatches
- NeuroNER API → Dataset loader: loads data
- NeuroNER API → Training and prediction: starts training
- NeuroNER API → Entity LSTM: builds model
- NeuroNER API → Pretrained models: fetches model
- NeuroNER API → Pretrained embeddings: loads vectors
- Dataset loader → BRAT to CoNLL: parses BRAT
- BRAT to CoNLL → Tokenizer services: requests tokenization
- i2b2 XML converter → CoNLL to BRAT: writes annotations
- Training and prediction → Entity LSTM: trains and predicts
- Training and prediction → Metrics and plots: evaluates labels
- Training and prediction → Prediction outputs: writes predictions
- NeuroNER API → CoNLL to BRAT: exports BRAT
- Model sharing prep → Pretrained models: prepares artifacts