Validation of sleep-based actigraphy machine learning models for prediction of preterm birth

This is the code for our paper, Validation of sleep-based actigraphy machine learning models for prediction of preterm birth (in submission).

Usage

Requirements are outlined in requirements.txt and can be installed via pip install -r requirements.txt. Data is not included, but if you wish to run your own survey data and/or summarized actigraphy data, references can be updated in the conf/conf.yaml file.

This project relies on the Hydra configuration system, with conf/conf.yaml containing the majority of the configuration. conf/ablation, conf/model contain the configurations for the ablation studies and model training, respectively. conf/names is used to map internal names to readable names. You will need to add a paths/root.yaml file pointing to the base of your data directory.

Training is done with python train.py, here are several example commands:

python train.py model=LinearSVCClassifier # Run linear SVC, base seed, all features
python train.py -m seed=175,4325,5132,09645,10239,23425,024248,59875,62345,657095 ablation=all_fts,no_sts,no_tabular model=glob(*) # All main models, with ablation studies
python train.py -m model=glob(*) ablation=all_nulliparous # One seed, all models, all nulliparous patients

Evaluation can then be done with python evaluate_all.py and plotting python plot.py.