This repository was developed to test adn compare different solution approaches (policies) for solving a Dynamic and Stochastic Inventory Management problem in a Blood Management context. It includes a Simulator based on Monte Carlo sampling and a set of different policies available for comparisson. For more details about the problem details and policies, please refer to How to solve Dynamic Resource Allocation problems using RL blog post.
This project was inspired by Warren Powell's Approximate Dynamic Programing book and the wbpowell328/stochastic-optimization repository.
This project uses Python 3.9 and all required packages can be found at the requirements.txt file.
blood-management-driver.py is the starting point of the project and drives the execution the simulation execution. The following params can use to trigger the execution:
"policies": ["myopic", "adp"], # Available policies: "myopic", "adp", and "basic"
"test_seed": 7383, # Random seed for reproducibility
"test_simulations": 100, # Number of simulations to run
"verbose": True, # Prints all scenarios' performance (reward and gap to perfect solution)
"scenarios_to_visualize": 0, # Number of scenarios to be graph and exported to the `figures` folder.
"instance_name": "epoch_15_age_3" # Instance name
The instances are loaded or generated within the (parameters.py)[parameters.py] file.
In case an instance named as the instance_name's value already exists on the
(instances/)[instances] folder, it will be loaded. Otherwise, a new instance file
will be created within the (instances/)[instances] folder. This new instance will use the
default params defined at the (parameters.py)[parameters.py] file.
Feel free to change the params to generate new instances.