PK Table Multilabel Classification Model
Model classes can be found:
Dataloaders can be found:
data_loaders/table_data_loaders.py
data_loaders/bow_data_loaders.py
data_loaders/kfold_CV_data_loaders.py
data_loaders/bootstrap_data_loaders.py
To initially check model is working (check overfit to small sample):
scripts/tableclass_debugFFNN.py
scripts/tableclass_debugBOW.py
scripts/tableclass_debugCNN.py
use with corresponding model config
To run model training and validation:
scripts/tableclass_trainFFNN.py
scripts/tableclass_trainBOW.py
scripts/tableclass_trainCNN.py
To run model testing:
scripts/tableclass_evaluateFFNN.py
scripts/tableclass_evaluateBOW.py
scripts/tableclass_evaluateCNN.py
To run k-fold cross validation:
To run Bootstrap:
scripts/FFNN_bootstrap.py
To run TensorBoard:
cd data
tensorboard --logdir=runs --bind_all
To plot:
#output of kfold cross validation:
scripts/plot_kfold.py
#loss and f1:
scripts/plot_loss_f1.py
#To plot sensitivity analysis:
scripts/plot_sens_analysis.py
To review labels vs model predictions (using Prodigy Software) and make corrections:
prodigy review final-test-qual-assessment labels,predictions -v choice
#get reviewed database out as JSONL file
prodigy db-out test-corrections > ./data/train-test-val/test-corrected.jsonl