Source code

Directory structure

• ./constants/: Contains constants used in the code, such as paths to datasets, glossaries, name mappings, etc.

• ./evaluation/: Contains code for evaluation, including generation of tables and plots, and computation of statistical tests.

• ./experiments/: Contains experiments for each method:

• ./experiments/rf.py: Random Forest
• ./experiments/bilstm.py: BiLSTM

• ./experiments/bert.py: ClinicalBERT architectures

• ./extensions/: Contains extensions to Python Baal and Transformers libraries.

• ./features/: Contains implementations of input features for RF and BiLSTM methods, and input representations for Clinical BERT and Paired Clinical BERT.

• ./ml_models/: Implementation of the different machine learning (ML) methods.

• ./models/: Contains data models for preprocessing and feature generation.

• ./preprocessing/: Contains preprocessing scripts.

• ./re_datasets/: Datasets factory for BiLSTM and BERT models, creating a Hugging Face (HF) Dataset.

• ./scripts/: Bash scripts for running experiments on the GPU cluster.

• ./training/: Contains trainers for each ML method and common training resources.

• ./config/: Configuration of the logging of results to Neptune.ai.

• ./nlp_pipeline.py: NLP Spacy pipeline.

• ./utils.py: Helper functions used throughout the code.

• ./vocabulary.py: Vocabulary module, representing mapping between tokens and indices.

Most of the subdirectories contain a more detailed README file.