Pass embedding matrix as parameter during MetaCAT training during cross-validation. I forgot to add this previously, and it significantly increases scores:
From cross-validation, gather results per example and save them in a new result file, results/bilstm_predictions_cv.csv.gz
Save scores from cross validation in new result file, results/bilstm_scores_cv.csv.gz
Add functioning MedCAT config file and run MedCAT in notebook 04_medcat_usage.ipynb with a recent concept database. SARS-CoV-2-infectie is now correctly recognized and linked, and the negation is also correctly identified!
Non-functional changes:
Add markdown documentation in many notebooks.
Merge the notebooks that create and evaluate the model based on the whole DCC set.
Attempt to improve naming and ordering of notebooks.
Remove everything from README.md that was not used.
Functional changes:
results/bilstm_predictions_cv.csv.gz
results/bilstm_scores_cv.csv.gz
04_medcat_usage.ipynb
with a recent concept database.SARS-CoV-2-infectie
is now correctly recognized and linked, and the negation is also correctly identified!Non-functional changes: