This repository contains a diabetic patient data simulator following the structure of an open-ai gym. It is intended to evaluate reinforcement learning algorithms.
For more information on how this gym can be used in applied reinforcement learning research, see our blog post on personalized dosing.
Install this package via pip:
python setup.py install
This simulator was inspired by the following work:
- Maintain Glucose in Type-I Diabetic
- simglucose, an implementation of the FDA-approved 2008 version UVA/Padova Simulator
This simulator is based on an expanded version of the Bergman minimal model, which includes meal disturbances. The underlying mathematical representation of this model was first developed by John D. Hedengren.
The goal is to keep glucose levels at a tolerable level in Type-1 diabetic patients. This process can be controlled using remote insulin uptake.
For additional details on this gym, see dosing_rl_gym/resources/Diabetic Background.ipynb.