Parsing Data:

• Place ELSA data in directory that the repository is in
• Execute Data_Parser/create_elsa_data.sh
• This reads the ELSA data, cleans it up, and splits it into three files for training, validation, and testing
• Also creates mean_deficits.txt and std_deficits.txt, used for calculating relative RMSE
• Execute data_info.py to create a file with information about the data

Model:

• located in /DJIN_Model
• model.py: main code for model
• diagonal_func.py: contains N neural networks for each deficit
• dynamics.py: calculates dynamics at a particular time step
• solver.py: uses dynamics on every time set
• loss.py: contains functions for calculating loss
• memory_model.py: contains nueral net used for calculating inital h for survival RNN
• vae_flow.py: contains variational autoencoder (VAE) for imputing data
• realnvp_flow.py: contains nvp flows used in VAE

Training:

• generate averages and standard deviations with population_average.py and population_std.py
• Execute train.py with a job_id
• Optionally set the hyperparameters, which are output to /Output
• Outputs trained parameters to /Parameters

Predicting:

• execute predict.py with a job_id and an epoch
• generates file with survival trajectories, used for c-index, brier score, and d-callibration
• generates file with mean trajectories for deficits, used for longitudinal predictions

Comparison model:

• located in /Comparison_model
• execute longitudinal.py to generate mean trajectories for deficits
• execute survival.py to generate survival trajectories

Latent model:

• located in /Alternate_model
• trained with train_full.py, predictions made with predict_full.py
• option to specify N
• generate variables to be used with generate_variables.py, orders variables by the amount they change in the data
• can also manually create variables.txt in /Data, height, bmi, ethnicity, and sex must be the last four variables
• create population averages and standard deviations with population_average_latent.py and population_std_latent.py