This repository contains files to build, train and test a deep-learning prediction model.

Interested scholars can access the data by contacting Nemours Biomedical Research Informatics Center and signing a data use agreement.

• cohort_selection_2.ipynb is the file used to select final cohort for descriptive analysis and prediction model training

• cohort_prelim is the file used to run descriptive analysis on the selected cohort.

• dl_train_sig_obs.py consists of code to create batches of data according to batch_size and create, train and test the model. Below are the function details -

  • create_model used to call mimic_model_sig_obs.py file to create and initiate the model architecture
  • decXY create input vector for decoder
  • encXY create input vector for encoder
  • dl_train calls all internal functions to train the model
  • dl_test calls all internal functions to test the model
  • model_test test the model to collect performance metrics
  • test_read reads the files created by model_test
  • display_output used to display the output of test results
  • model_test_full test the model for feature importance interpretation
  • interpret create files with results of feature importance interpretation and save the files
  • interpret_read reads the files creates by interpret function and calls display_contri function to display the output
  • display_contri used to display feature importance

• dl_train_sig_obs_geo.py consists of code to create batches of data for temporal and geographic validation of the model.

• mimic_model_sig_obs.py consist of different model architectures.

• evaluation.py consists of class to perform evaluation of results obtained from models.

• fairness.py consists of code to perform fairness evaluation.

• parameters.py consists of list of hyperparameters to be defined for model training.

• ./saved_models consists of models saved during training.