HeartPole is a simple Gymnasium for benchmarking Reinforcement Learning algorithms. Think CartPole for healthcare. It is meant as a first test for reinforcement learning techniques aimed at the Healthcare domain, before moving on to actual clinical data like MIMIC III
This project includes a tutorial on creating your own Gymnasium environment, using HeartPole as an example, as well as a writeup of experimental results, submitted to HEALTHINF conference.
You are a developer working on an important project. You want to maximize your productivity, but not at the expense of health - any serious health issue will negate all the effects of increased productivity.
Action space is discrete with 4 possible actions:
[do_nothing, drink_coffee, drink_beer, sleep]
Observation space is continious, with 6 dimensions:
['alertness', 'hypertension', 'intoxication',
'time_since_slept', 'time_elapsed', 'work_done']
Your productivity mainly depends on alertness, you get a reward of 1 for every unit of work done and a reward of -100 if you suffer a heart attack.
Install with
pip install heartpole
Create an environment with
from heartpole import HeartPole
env = HeartPole()
and use your favourite reinforcement learning algorithm, for example, from keras-rl