Dataset for a reinforcement learning application involving glucose homeostasis in the ICU, using the MIMIC-III Clinical Dataset
Data curation repository based on 4yr_project_glucose
by Zhiyao Luo.
Update: This project was presented at the Institute of Biomedical Engineering, University of Oxford on Sep 2, 2024. You can find the presentation slides here.
To run this project locally, the following is required:
Note: The RxNorm and RxClass APIs used inside RxNav-in-a-Box must be run locally. Read more at Classifying prescriptions.
You can read documentation articles without running the project locally. Essential plots utilised in the literature are pushed to enable this.
From the RxNav-in-a-Box README.txt:
- "12 gigabytes of memory to devote to a container platform (e.g., Docker)
- 100 gigabytes of disk space
- Docker Desktop,
or another OCI-compatible platform (in which case you
may take the included docker-compose.yml file as an example)."
From
the mimic-code
repository:
"Loading the data into a PostgreSQL database requires around ~47 GB of space. The addition of [optional] indexes adds another 26 GB. You will likely want to reserve 100 GB for the entire database."
If running both the RxNav-in-a-Box and MIMIC-III databases locally, ensure that you have enough disk space and memory:
Docker Desktop is recommended for running RxNav-in-a-Box locally.
.zip
file containing the dataset Curated Data for Describing Blood Glucose Management in the.zip
file containing the LOINC database from loinc.org and uncompress.env
file with your database credentials.pip install -r requirements.txt
..env
.curation
module inside your virtual environment using python -m curation
.name or flags | type | default | description |
---|---|---|---|
-l , --log-level
|
str |
warning |
The log level. |
-m , --max-identifier-count
|
int |
-1 |
The maximum number of unique ICU stays identifiers. Any number less than or equal to zero will not limit the number of identifiers, and all will be used. |
-c , --chunk_size
|
int |
1000 |
The chunk size to use when querying the database. |
Tahmid Azam, [email protected]