This is the code repository for the meds_reader ML4H paper submission.
It consists of four main components, each corresponding to a folder:
-
meds_reader: Our meds_reader software package. Install this using pip install -e .
-
reference_implementaion: Reference implementations for length of stay labeling and featurization for both PyHealth and MIMIC-IV-Data Pipeline. Note the README for running against MIMIC-IV Data Pipeline.
-
conversion: Scripts for converting from PyHealth to MIMIC-IV Data Pipeline to MEDS format for use with meds_reader.
Make sure to install PyHealth from their Github (https://github.com/sunlabuiuc/PyHealth). Likewise, install MIMIC-IV Data Pipeline using their instructions (https://github.com/healthylaife/MIMIC-IV-Data-Pipeline).
After conversion, make sure to run meds_reader_convert on both of the result folders
meds_reader_convert datasets/pyhealth_meds datasets/pyhealth_meds_reader --num_threads 10
meds_reader_convert datasets/pipeline_meds datasets/pipeline_meds_reader --num_threads 10- reimplementation: Reimplementations of length of stay labeling and featurization using meds_reader. Make sure to run the conversion scripts as well as meds_reader_convert before running.
In order to run any of these scripts, you must first create a "datasets" folder with downloads of MIMIC-III and MIMIC-IV.