ckd_progression
===============

Predict outcomes related to Chronic Kidney Disease. For more info, see:

https://github.com/clinicalml/deepDiagnosis

To run the full pipeline from cohort construction to prediction on the fake data:

python ckd_progression.py tests/test_config.yaml tests/test_data_paths.yaml tests/test_stats.yaml tests/kidney_disease/

The program takes as input (1) a YAML file (e.g. tests/test_data_paths.yaml) with the paths to Python shelve databases storing the raw data and text files storing lists of patient IDs and medical codes, (2) a YAML file (e.g. tests/test_stats.yaml) that defines various statistics used to build a cohort and assign outcomes, and (3) a path to a directory in which to store the output.  

The pipeline makes use of the following programs:

- patient_stats.py: Calculate statistics on groups of patients
- build_training_data.py: Build training examples by sliding a window over the lab data
- features.py: Extract a set of lab, diagnosis, and prescription features for each training example and randomly divide the examples into a training, validation and test set
- predict.py: Train an L2-regularized logistic regression, an L1-regularized logistic regression, a Random Forest and a Convolutional Neural Network, use cross-validation to select the hyperparameters and calculate the AUC on the test set for the best models

To run tests from the command line:

nosetests