HHMM

Hierarchical Hidden Markov Model

Code

1. Experiment batch script:
   • Batch job script for model training: Experiments/EM_bootstrapt_sbatch
2. Clinical applications:
   • Codes in clinical_applications is used for risk stratification. Basically, it compares our HHMM based method with baseline method for risk stratification in terms of KM curve metric.
3. Model comparison:
   • This compare our model with naive HMM model as well as a batch of deep learning based models in terms of model prediction performance.
   • In particular,
     • Code for HHMM is in Model_comparison/EM_hierechical_prediction.py.
     • Code for HMM is in Model_comparison/EM_prediction.py.
     • Codes for deep learning methods are in Model_comparison/deep_learning.
4. Model inference:
   • We have two versions of inference code.
     • One for python2 in Model_inference/python2/HHMM_inference.py.
     • The other for python3 in Model_inference/python3/EM_hierarchical_paralleled.py with the bootstrap version in Model_inference/python3/EM_hierarchical_paralleled_bootstrap.py.
5. Model prediction:
   • Codes in this folder are used to predict the score for three screening results based on historical records.
   • In particular,
     • Prediction code for HHMM is in Model_prediction/EM_hierechical_prediction_SR.py.
     • Prediction code for HMM is in Model_prediction/EM_prediction_SR.py.
6. Model validation:
   • Code to plot Kaplan Meier curve given HMM and HHMM models is provided in Model_validation/model_validation.py.
7. Simulation:
   • To illustrate the correctness of inference code, we conduct synthetic experiments in which date are generated from HHMM. All simulation code and corresponding inference codes are provided in Simulation.
8. Survey_applications:
   • Codes for HHMM model based application with survey data are provided in Survey_applications.
9. Others:
   • We provided the data analysis codes and ultility codes in Others.

Data

Data are available for inquiry.